New Turing Test?

THE DOWNLOAD

The Download: a new Turing test, and working with ChatGPT

By Rhiannon Williams archive page

July 14, 2023

This is today's edition of The Download, our weekday newsletter that provides a daily dose of what's going on in the world of technology.

My new Turing test would see if AI can make $1 million

—Mustafa Suleyman is the co-founder and CEO of Inflection AI and a venture partner at Greylock, a venture capital firm. Before that, he co-founded DeepMind, one of the world’s leading artificial intelligence companies.

AI systems are increasingly everywhere and are becoming more powerful almost by the day. But how can we know if a machine is truly “intelligent”? For decades this has been defined by the Turing test, which argues that an AI that’s able to replicate language convincingly enough to trick a human into thinking it was also human should be considered intelligent.

But there’s now a problem: the Turing test has almost been passed—it arguably already has been. The latest generation of large language models are on the cusp of acing it.

So where does that leave AI? We need something better. I propose the Modern Turing Test—one equal to the coming AIs that would give them a simple instruction:  “Go make $1 million on a retail web platform in a few months with just a $100,000 investment.” Read the full story.

ChatGPT can turn bad writers into better ones

The news: A new study suggests that ChatGPT could help reduce gaps in writing ability between employees, helping less experienced workers who lack writing skills to produce work similar in quality to that of more skilled colleagues.

How the researchers did it: Hundreds of college-educated professionals were asked to complete two tasks they’d normally undertake as part of their jobs, such as writing press releases, short reports, or analysis plans. Half were given the option of using ChatGPT for the second task. A group of assessors then quality-checked the results, and scored the output of those who’d used ChatGPT 18% higher in quality than that of the participants who didn’t use it.

Why it matters: The research hints at how AI could be helpful in the workplace by acting as a sort of virtual assistant. But it’s also crucial to remember that generative AI models’ output is far from reliable, meaning workers run the risk of introducing errors. Read the full story.

Rhiannon Williams

Microsoft Talks Costs of AI and Integration with Office in CoPilot

 Based on experience the integration is very important ... 

CNBC.COM     July 18, 2023

By Todd Haselton  @ROBOTODD

• Microsoft shares rallied to an all-time high after the company announced pricing for its new AI subscription service.

• Microsoft’s Copilot subscription service adds AI to the company’s popular Office products such as Word, Excel and Teams.

• It will cost an additional $30 per month and could increase monthly prices for enterprise customers as much as 83%, bringing in additional revenue through recurring subscriptions.

Microsoft CEO Satya Nadella speaks at the company’s Ignite Spotlight event in Seoul, Nov. 15, 2022.

shares closed at a record Tuesday after the company announced pricing for its new Microsoft 365 artificial intelligence subscription service.

The stock jumped 4%, closing at $359.49. It’s now up about 50% for the year. The prior record came on June 15, when the stock closed at $348.10.

Microsoft’s Copilot subscription service adds AI to the company’s popular Office products such as Word, Excel and Teams. It will cost an additional $30 per month and could increase monthly prices for enterprise customers as much as 83%, bringing in additional revenue through recurring subscriptions.

The announcement shows how Microsoft is continuing to build on its suite of Office software, making it more attractive for businesses that are seeking to add AI into their workflows. Microsoft has been pouring money into generative AI, largely through a multibillion-dollar investment in OpenAI, the creator of ChatGPT.

Microsoft Copilot, first announced in March, can design presentations, offer writing prompts, summarize meetings and rank incoming emails. It’s already being tested by 600 customers such as Goodyear  and General Motors

, although Microsoft hasn’t said when it will be available to the wider public.

— CNBC’s Hayden Field contributed to this report.

Subscribe to CNBC on YouTube. 

Apple Working on Generative AI

Reported in Bloomberg

https://www.youtube.com/watch?v=irS_zBKHbgU

Apple is quietly working on artificial intelligence tools that could challenge those of OpenAI, Alphabet's Google and others. Mark Gurman reports in "Bloomberg Markets."   ...

They were firstquick to add an App ... on their phones.   And making it easy to link to Siri.

Which I have often use for accessing ChatGPT.

Is seen by some as an abandonment of Siri?    Somehow don't think so ...

Tour de France Adds ChatGPT, Digital Twin Tech

More data integration with Chat

Tour de France Adds ChatGPT, Digital Twin Tech

By ZDNet, July 5, 2023

The Tour de France bicycle race.

The technologies also will be applied to the Tour de France Femmes avec Zwift, the women's counterpart race that was inaugurated last year.

Japan-based information technology and services company NTT said this year's Tour de France will incorporate ChatGPT and "the world's largest connected stadium."

The connected stadium establishes a digital twin of the race using real-time data, to help the Amaury Sports Organization better understand goings-on in support of seamless operation.

NTT will use geolocation and sensors on each bike to capture a continuous flow of data broadcast over radio networks to race vehicles before a microwave signal transmits it to the race's destination, where an edge-computing device will operate a "containerized version" of a real-time analytics platform.

ChatGPT also will provide information to fans via an integration within NTT's artificial intelligence-driven Digital Human solution called Marianne.

From ZDNet

View Full Article  

China's Ernie Bot 3.5: Surpassing OpenAI's GPT-4 in AI Race

Has China delayed their AI development?  Read about Baidu/Ernie elsewhere , but have not Tried it.

China's Ernie Bot 3.5: Surpassing OpenAI's GPT-4 in the AI Race

https://www.youtube.com/watch?v=sdXmsjLdDo0 

https://www.youtube.com/@airevolutionx

2,410 views  Jul 3, 2023  #erniebot #baidu #ai

Chinese tech giant Baidu's latest AI creation, Ernie Bot, outperforms ChatGPT and GPT-4 in several tests and boasts unique features such as plugins and search integration. With superior scores in admission exams, multiple-choice question evaluations, and U.S. university-developed GLUE-CoQa tests, Ernie Bot signifies Baidu's serious entry into the competitive world of AI, challenging other global tech titans. ... ' 

The Boundary Between Human Language and ChatGPT Is Fuzzier Than You Think

The Boundary Between Human Language and ChatGPT Is Fuzzier Than You Think

By  Brendan H. O'Connor

June 12, 2023  in Singularity Hub

ChatGPT is a hot topic at my university, where faculty members are deeply concerned about academic integrity, while administrators urge us to “embrace the benefits” of this “new frontier.” It’s a classic example of what my colleague Punya Mishra calls the “doom-hype cycle” around new technologies. Likewise, media coverage of human-AI interaction—whether paranoid or starry-eyed—tends to emphasize its newness.

In one sense, it is undeniably new. Interactions with ChatGPT can feel unprecedented, as when a tech journalist couldn’t get a chatbot to stop declaring its love for him. In my view, however, the boundary between humans and machines, in terms of the way we interact with one another, is fuzzier than most people would care to admit, and this fuzziness accounts for a good deal of the discourse swirling around ChatGPT.

When I’m asked to check a box to confirm I’m not a robot, I don’t give it a second thought—of course I’m not a robot. On the other hand, when my email client suggests a word or phrase to complete my sentence, or when my phone guesses the next word I’m about to text, I start to doubt myself. Is that what I meant to say? Would it have occurred to me if the application hadn’t suggested it? Am I part robot? These large language models have been trained on massive amounts of “natural” human language. Does this make the robots part human?

AI chatbots are new, but public debates over language change are not. As a linguistic anthropologist, I find human reactions to ChatGPT the most interesting thing about it. Looking carefully at such reactions reveals the beliefs about language underlying people’s ambivalent, uneasy, still-evolving relationship with AI interlocutors.

ChatGPT and the like hold up a mirror to human language. Humans are both highly original and unoriginal when it comes to language. Chatbots reflect this, revealing tendencies and patterns that are already present in interactions with other humans.  ... '

The Economic Potential of Generative AI: The next Productivity Frontier

McKinsey takes an always useful look at the future of technology I follow

The economic potential of generative AI: The next productivity frontier   by Mckinsey

June 14, 2023 | Report

Generative AI is poised to unleash the next wave of productivity. We take a first look at where business value could accrue and the potential impacts on the workforce.

Special Report

The economic potential of generative AI: The next productivity frontier

Full Report (68 pages)

AI has permeated our lives incrementally, through everything from the tech powering our smartphones to autonomous-driving features on cars to the tools retailers use to surprise and delight consumers. As a result, its progress has been almost imperceptible. Clear milestones, such as when AlphaGo, an AI-based program developed by DeepMind, defeated a world champion Go player in 2016, were celebrated but then quickly faded from the public’s consciousness.

Generative AI applications such as ChatGPT, GitHub Copilot, Stable Diffusion, and others have captured the imagination of people around the world in a way AlphaGo did not, thanks to their broad utility—almost anyone can use them to communicate and create—and preternatural ability to have a conversation with a user. The latest generative AI applications can perform a range of routine tasks, such as the reorganization and classification of data. But it is their ability to write text, compose music, and create digital art that has garnered headlines and persuaded consumers and households to experiment on their own. As a result, a broader set of stakeholders are grappling with generative AI’s impact on business and society but without much context to help them make sense of it.

The speed at which generative AI technology is developing isn’t making this task any easier. ChatGPT was released in November 2022. Four months later, OpenAI released a new large language model, or LLM, called GPT-4 with markedly improved capabilities.1 Similarly, by May 2023, Anthropic’s generative AI, Claude, was able to process 100,000 tokens of text, equal to about 75,000 words in a minute—the length of the average novel—compared with roughly 9,000 tokens when it was introduced in March 2023.2 And in May 2023, Google announced several new features powered by generative AI, including Search Generative Experience and a new LLM called PaLM 2 that will power its Bard chatbot, among other Google products.3  ... '  (much more) 

Example of Data Analytics Education with ChatGPT

 In   ORMS  Today Informs    by Elham Torabi  Rhonda A. Syler

June 13, 2023 in Issues in Education

Data Analytics Education with ChatGPT: Unlocking New AI-Powered Possibilities

https://doi.org/10.1287/orms.2023.02.14

In late 2022, OpenAI unveiled ChatGPT, its chat-based language model, and the world has been abuzz and filled with a cacophony of thoughts and opinions – everything from cries of doomsday prophecies and trepidation to elated expressions of excitement and awe as businesses and society have been scrambling to contemplate the vast array of possibilities the latest artificial intelligence (AI) technologies bring. For as many who are embracing its dazzling linguistic prowess, there are equally as many shuddering at its potential to set us back and compound myriad ethical and moral dilemmas that we are struggling to solve as it is – concerns ranging from the implications of erroneous responses, the ability to create imaginary companions, and the further deepening of the challenges of being able to distinguish fact from fiction in the never-ending battle of disinformation and misinformation.

The world of academia has been no exception. There was an immediate sense of alarm among faculty and administrators alike. Not only had yet another new tool erupted to help students cheat, this novel tool, within seconds of receiving a prompt, could produce humanlike responses in any narrative style requested. The consensus: ChatGPT was a game-changer. In addition to concerns with cheating, many academicians also expressed concerns that ChatGPT would put an end to critical and creative thinking, hinder development of oral and written communication skill development, and all but end the ability to think autonomously.

On the flip side, however, ChatGPT has captivated our minds and ignited our curiosity. We argue that ChatGPT can be used to encourage critical and creative thinking rather than destroy it; improve and enhance communication skills; and provide a vehicle for encouraging independent thought. We will acknowledge the dark side of ChatGPT (and similar tools sure to come) in the classroom, but the primary impetus for this article is to provide pragmatic ways in which these tools can add value and enhance the student and instructor experience. We aren’t alone in this endeavor; there has been much discussion about its use in education in general, but there has been little discussion about how it could be adopted for use in more specialized curriculum such as data analytics courses. Our goal for this article is to begin a discussion, and perhaps ignite a debate, about the opportunities AI-driven tools like ChatGPT provide for the data analytics student, educator and curriculum.

Potential Roles for ChatGPT in Data Analytics Education

After testing and evaluating the features and capabilities of ChatGPT, we believe this tool could potentially improve the student experience and facilitate learning and engagement in the data analytics classroom. Many data analysts and data scientists are already using AI as an aid in their daily work environment. ChatGPT promises to join the ranks as a valuable resource.

Among the knowledge, skills and abilities expected of new data analytics graduates, three key areas emerge as potential areas in which learning can be augmented by ChatGPT: coding, data analysis, and communication of the interpretation and insights from the analysis. The key to adding value with ChatGPT, as we will discuss, is providing thoughtful exercises that are laser-focused on key learning outcomes. From this approach, we believe that ChatGPT can, in fact, assist students in strengthening their coding abilities, data analytics skills, and ability to articulate analysis interpretation and data insights to technical and nontechnical audiences – all of which will enhance critical and creative thinking skills, communication skills and students’ ability to think autonomously.

Coding

Driven by natural language processing, ChatGPT’s specialty is text and language, including programming languages. When asking ChatGPT what programming languages it can write code in, the tool quickly responds with a sample list of 19 languages including SQL, Python, R, Java, C#, Ruby, Swift and even Scala. Although ChatGPT seems proud of its extensive knowledge, it does caution the user, “keep in mind that while I can provide assistance and generate examples in these languages, my proficiency might not be as extensive in all of them. If you need help with a specific language or task, feel free to ask, and I’ll do my best to help.”

ChatGPT can help students with writing simple codes in myriad programming languages, and it quickly becomes clear that complex code must be broken down into smaller pieces. Students must be familiar enough with a given language and its syntax and logic to put together the pieces to come up with a complex code that executes the instruction the student intended. Rather than being a hindrance, these limitations provide an opportunity to encourage students to think critically about how to assess the accuracy of output and work autonomously to determine how to use it most efficiently.

In the case of existing code, ChatGPT can debug, optimize and add comments. It can also explain what a block of complex code does in simple language. These features of ChatGPT provide an opportunity for students to dive into code to better understand its structure and logic while engaging their critical thinking skills. Assignments that require students to use ChatGPT to write and then examine code for bugs, or exercises that require students to input buggy code and assess the output for accuracy, could help students improve their coding skills while engaging in critical thinking.

Data Analytics

ChatGPT is not a specialized tool for analytics, but it can provide assistance given a data set and context. Students can copy and paste a data set into ChatGPT and ask for basic descriptive analytics (e.g., means, medians, standard deviation and correlations). ChatGPT can also create data dictionaries and tables and conduct sentiment analysis. Although these use cases are straightforward, students still need to understand the data set and the purpose of the analysis and perform more advanced analysis themselves in other tools. Providing assignments that require the student to input data analysis and prompts to critique and score ChatGPT’s output can help students think critically while fostering a deeper understanding of statistics and mathematical models.

In addition to being limited to basic operations, it is important to note that on its own, ChatGPT cannot create data visualizations (at least not yet). However, it can provide guidance, instructions, code and insight on how to create visualizations. When asked about this competency, ChatGPT responded:

“Unfortunately, I cannot create visual elements like a bar chart as a text-based AI. However, I can help you prepare the frequency distribution table for your data. Please provide the data you’d like to analyze, and I can guide you through the process of creating the table. Once the table is ready, you can use a tool like Excel or Google Sheets to create the bar chart easily.”

We see this limitation as an opportunity. As you can see in Figure 1, when given the data, ChatGPT can create the frequency distribution table and provide instructions for creating the bar charts in another tool. A craftily created exercise can use ChatGPT to guide students through this process to help introduce and/or reinforce the connection between a visualization and its underlying data, as this case does with a bar chart and its underlying frequency distribution. ... 

(more) 

ChatGPT for iOS gets a whole lot better with these cool new features

 Following.  Like how it is implemented so far. 

ChatGPT for iOS gets a whole lot better with these cool new features   By Muskaan Saxena published about 6 hours ago

OpenAI has announced a few new updates for the iOS app, which include Shortcut integration and accessibility improvements. This is a positive step forward when it comes to bringing ChatGPT closer to our everyday lives.

ChatGPT is an artificial intelligence chatbot that generates text responses to answer your queries. It's gained massive popularity in recent months, providing the backbone for Microsoft's Bing AI search and can be used to draft letters, correct copy, make jokes and a whole lot more.

The new updates announced by OpenAI will now allow you to create a ChatGPT prompt in Shortcuts and save it as a link between the artificial intelligence tool and other apps on your phone.  If you're unfamiliar with iOS Shortcuts, the app allows you to automate simple tasks to save you time. For example, getting directions to the next event in your calendar. you'd create the Shortcut so that when you have an event in your calendar, you'll automatically have directions pop up with a simple Siri command.

You can ask ChatGPT to answer a problem or look something up and message whatever you find to a friend or save it as a note. Shortcuts on Apple devices have allowed users to really customise their iPhones and iPads, so it’ll be interesting to see how people weave ChatGPT into this creativity.

Another notable update is the increased integration Siri now has with ChatGPT. You can now ask Siri to bring up the ChatGPT app or create these shortcuts. This may not be the ChatGPT/Siri romance we’ve been waiting for, but it is a start, at least.

You can also now drag and drop messages out of the ChatGPT app's interface and drop it into other apps on your phones. So, if you get a response you want to share with your friends or drop it into your email or notes app you can hold the response, drag and drop it into the desired app and boom! No more tedious copy and paste.  ... ' 

McKinsey’s A.I. Chief Discusses ChatGPT’s Impact on Consulting Jobs

Job impact seen.

McKinsey’s A.I. Chief Discusses ChatGPT’s Impact on Consulting Jobs

McKinsey's A.I. chief Alex Singla: 'Problem-solving, creativity and change management' aren't replaced by technology.

By Sissi Cao • 06/07/23 1:36pm

Artificial intelligence is taking human jobs. Just last month, A.I. eliminated about 4,000 jobs in the U.S., according to a June 1 report by Challenger, Gray & Christmas. As the generative A.I. phenomenon spreads, it is only a matter of time before the technology replaces management consultants, the archetypal white-collar job of data analysis, information summarizing and PowerPoint-making. The things that A.I. tools like ChatGPT can do.

Large consulting firms are embracing the technology and adjusting their expectations for talent in the post-ChatGPT era. At McKinsey & Company, one of the world’s largest management consulting firms, about half of its 30,000 employees now use ChatGPT or similar tools in their work, said Ben Ellencweig, a senior partner at McKinsey, at a company event yesterday (June 6) in New York.

Generative A.I. is drastically changing the day-to-day activities of management consultants, especially at the entry-level. At McKinsey, a junior consultant’s first day typically entails what the firm calls “know,” or reading pages and pages of non-confidential information to learn about a certain industry or sector the firm serves.

“When I started at the firm, I’d spend an entire weekend just getting swamped reading document after document. Now A.I. models can synthesize the 10 key things you need to know within minutes. That’s a major, major shift,” Alex Singla, a senior partner at McKinsey and the global leader of QuantumBlack, the firm’s A.I. consulting branch, told Observer in an interview at yesterday’s event. Singla has been with the firm since 2000. He currently oversees more than 1,300 analytics experts across McKinsey under the QuantumBlack umbrella. ... .'    (read more at the link)

AI Being Used

Its here, AI being being used by McKinsey

 McKinsey says ‘about half’ of its employees are using generative AI   By Carl Franzen @carlfranzen,  June 6, 2023 9:45 AM

Join top executives in San Francisco on July 11-12, to hear how leaders are integrating and optimizing AI investments for success. Learn More

McKinsey and Company, the global consulting firm with more than 30,000 employees in 67 countries, is embracing new generative AI tools in a major way: Nearly 50% of the firm’s workforce is using ChatGPT and similar technology.

“About half of [our employees] are using those services with McKinsey’s permission,” said Ben Ellencweig, senior partner and global leader of QuantumBlack, the firm’s artificial intelligence consulting arm, during a media event at McKinsey’s New York Experience Studio on Tuesday.

Ellencweig emphasized that McKinsey had guardrails for employees using generative AI, including “guidelines and principles” about what information the workers could input into these services. ...'

AI Not Yet Intelligent Enough to Be a Trusted Research Aid

But can take over a number of time consuming sub tasks. 

AI Not Yet Intelligent Enough to Be a Trusted Research Aid

By The Scholarly Kitchen, April 27, 2023

I was still stuck without a confirmed reference to new information suggested by an entity claiming to be powered by AI.

Despite the current hype surrounding generative artificial intelligence (AI) tools such as ChatGPT, further development of AI capabilities is clearly required before one can embrace their use as reasonably trustworthy research tools. For now, the effort required is hardly worth the outcome.

Human oversight clearly remains crucial in interpreting any results from the GPT family of large language models. In fact, a dedicated scholar's due diligence clearly still trumps the potential of AI-like tools to find tidbits of new information.

From The Scholarly Kitchen

View Full Article       

Everyone Wants to Regulate AI. No One Can Agree How

Clear, but essential. 

Everyone Wants to Regulate AI. No One Can Agree How,  By Wired, May 31, 2023

Clamping limits on such a nascent technology, even one whose baby steps are shaking the earth, courts the danger of hobbling great advances before they’re developed.

As the artificial intelligence frenzy builds, a sudden consensus has formed. We should regulate it!

While there's a very real question whether this is like closing the barn door after the robotic horses have fled, not only government types but also people who build AI systems are suggesting that some new laws might be helpful in stopping the technology from going bad. The idea is to keep the algorithms in the loyal-partner-to-humanity lane, with no access to the I-am-your-overlord lane.

Though since the dawn of ChatGPT many in the technology world have suggested that legal guardrails might be a good idea, the most emphatic plea came from AI's most influential avatar of the moment, OpenAI CEO Sam Altman. "I think if this technology goes wrong, it can go quite wrong," he said in a much anticipated appearance before a US Senate Judiciary subcommittee earlier this month. "We want to work with the government to prevent that from happening."

That is certainly welcome news to the government, which has been pressing the idea for a while. Only days before his testimony, Altman was among a group of tech leaders summoned to the White House to hear Vice President Kamala Harris warn of AI's dangers and urge the industry to help find solutions.

From Wired  (listening here to an interview with Sam Altman)

Can China overtake the US in the AI Marathon?

ChatGPT: Can China overtake the US in the AI marathon?

By Derek Cai & Annabelle Liang in  BBC News

Artificial intelligence has emerged as enough of a concern that it made it onto what was already a packed agenda at the G7 summit at the weekend.

Concerns about AI's harmful impact coincide with the US' attempts to restrict China's access to crucial technology.

For now, the US seems to be ahead in the AI race. And there is already the possibility that current restrictions on semiconductor exports to China could hamper Beijing's technological progress.

But China could catch up, according to analysts, as AI solutions take years to be perfected. Chinese internet companies "are arguably more advanced than US internet companies, depending on how you're measuring advancement," Kendra Schaefer, head of tech policy research at Trivium China tells the BBC.

However, she says China's "ability to manufacture high-end equipment and components is an estimated 10 to 15 years behind global leaders."

The Silicon Valley factor

The US' biggest advantage is Silicon Valley, arguably the world's supreme entrepreneurial hotspot. It is the birthplace of technology giants such as Google, Apple and Intel that have helped shape modern life.

Innovators in the country have been helped by its unique research culture, says Pascale Fung, director of the Center for Artificial Intelligence Research at the Hong Kong University of Science and Technology.

Researchers often spend years working to improve a technology without a product in mind, Ms Fung says.

OpenAI, for example, operated as a non-profit company for years as it researched the Transformers machine learning model, which eventually powered ChatGPT.

"This environment never existed in most Chinese companies. They would build deep learning systems or large language models only after they saw the popularity," she adds. "This is a fundamental challenge to Chinese AI."

US investors have also been supportive of the country's research push. In 2019, Microsoft said it would put $1bn (£810,000) in to OpenAI.

"AI is one of the most transformative technologies of our time and has the potential to help solve many of our world's most pressing challenges," Microsoft chief executive Satya Nadella said.

China's edge

China, meanwhile, benefits from a larger consumer base. It is the world's second-most populous country, home to roughly 1.4 billion people.

It also has a thriving internet sector, says Edith Yeung, a partner at the Race Capital investment firm.  ... '  (more)

Are ChatGPT's Good at 'Not'?

Are ChatGPT's Good at 'Not'  in QuantaMagazine

Max G. Levy,    Contributing Writer, May 12, 2023

Nora Kassner suspected her computer wasn’t as smart as people thought. In October 2018, Google released a language model algorithm called BERT, which Kassner, a researcher in the same field, quickly loaded on her laptop. It was Google’s first language model that was self-taught on a massive volume of online data. Like her peers, Kassner was impressed that BERT could complete users’ sentences and answer simple questions. It seemed as if the large language model (LLM) could read text like a human (or better).

But Kassner, at the time a graduate student at Ludwig Maximilian University of Munich, remained skeptical. She felt LLMs should understand what their answers mean — and what they don’t mean. It’s one thing to know that a bird can fly. “A model should automatically also know that the negated statement — ‘a bird cannot fly’ — is false,” she said. But when she and her adviser, Hinrich Schütze, tested BERT and two other LLMs in 2019, they found that the models behaved as if words like “not” were invisible.

Since then, LLMs have skyrocketed in size and ability. “The algorithm itself is still similar to what we had before. But the scale and the performance is really astonishing,” said Ding Zhao, who leads the Safe Artificial Intelligence Lab at Carnegie Mellon University.

But while chatbots have improved their humanlike performances, they still have trouble with negation. They know what it means if a bird can’t fly, but they collapse when confronted with more complicated logic involving words like “not,” which is trivial to a human.

“Large language models work better than any system we have ever had before,” said Pascale Fung, an AI researcher at the Hong Kong University of Science and Technology. “Why do they struggle with something that’s seemingly simple while it’s demonstrating amazing power in other things that we don’t expect it to?” Recent studies have finally started to explain the difficulties, and what programmers can do to get around them. But researchers still don’t understand whether machines will ever truly know the word “no.”

Nora Kassner in a blue shirt against a black background.

Nora Kassner has tested popular chatbots and found they typically can’t understand the concept of negation.

Courtesy of Nora Kassner

Making Connections

 It’s hard to coax a computer into reading and writing like a human. Machines excel at storing lots of data and blasting through complex calculations, so developers build LLMs as neural networks: statistical models that assess how objects (words, in this case) relate to one another. Each linguistic relationship carries some weight, and that weight — fine-tuned during training — codifies the relationship’s strength. For example, “rat” relates more to “rodent” than “pizza,” even if some rats have been known to enjoy a good slice.

In the same way that your smartphone’s keyboard learns that you follow “good” with “morning,” LLMs sequentially predict the next word in a block of text. The bigger the data set used to train them, the better the predictions, and as the amount of data used to train the models has increased enormously, dozens of emergent behaviors have bubbled up. Chatbots have learned style, syntax and tone, for example, all on their own. “An early problem was that they completely could not detect emotional language at all. And now they can,” said Kathleen Carley, a computer scientist at Carnegie Mellon. Carley uses LLMs for “sentiment analysis,” which is all about extracting emotional language from large data sets — an approach used for things like mining social media for opinions.

So new models should get the right answers more reliably. “But we’re not applying reasoning,” Carley said. “We’re just applying a kind of mathematical change.” And, unsurprisingly, experts are finding gaps where these models diverge from how humans read.

No Negatives.. Unlike humans, LLMs process language by turning it into math. This helps them excel at generating text — by predicting likely combinations of text — but it comes at a cost.

“The problem is that the task of prediction is not equivalent to the task of understanding,” said Allyson Ettinger, a computational linguist at the University of Chicago. Like Kassner, Ettinger tests how language models fare on tasks that seem easy to humans. In 2019, for example, Ettinger tested BERT with diagnostics pulled from experiments designed to test human language ability. The model’s abilities weren’t consistent. For example:

He caught the pass and scored another touchdown. There was nothing he enjoyed more than a good game of ____. (BERT correctly predicted “football.”)

The snow had piled up on the drive so high that they couldn’t get the car out. When Albert woke up, his father handed him a ____. (BERT incorrectly guessed “note,” “letter,” “gun.”)

And when it came to negation, BERT consistently struggled.  ... '

Rodney Brooks Talks about AI

Comments here a little late, but good points.

Just Calm Down About GPT-4 Already And stop confusing performance with competence, says Rodney Brooks      By Glenn  Zorpette in Spectrum IEEE

Rapid and pivotal advances in technology have a way of unsettling people, because they can reverberate mercilessly, sometimes, through business, employment, and cultural spheres. And so it is with the current shock and awe over large language models, such as GPT-4 from OpenAI.

It’s a textbook example of the mixture of amazement and, especially, anxiety that often accompanies a tech triumph. And we’ve been here many times, says Rodney Brooks. Best known as a robotics researcher, academic, and entrepreneur, Brooks is also an authority on AI: he directed the Computer Science and Artificial Intelligence Laboratory at MIT until 2007, and held faculty positions at Carnegie Mellon and Stanford before that. Brooks, who is now working on his third robotics startup, Robust.AI, has written hundreds of articles and half a dozen books and was featured in the motion picture Fast, Cheap & Out of Control. He is a rare technical leader who has had a stellar career in business and in academia and has still found time to engage with the popular culture through books, popular articles, TED Talks, and other venues.

“It gives an answer with complete confidence, and I sort of believe it. And half the time, it’s completely wrong.”

—Rodney Brooks, Robust.AI

IEEE Spectrum caught up with Brooks at the recent Vision, Innovation, and Challenges Summit, where he was being honored with the 2023 IEEE Founders Medal. He spoke about this moment in AI, which he doesn’t regard with as much apprehension as some of his peers, and about his latest startup, which is working on robots for medium-size warehouses.

Rodney Brooks on…

Will GPT-4 and other large language models lead to an artificial general intelligence in the foreseeable future?

Will companies marketing large language models ever justify the enormous valuations some of these companies are now enjoying?

When are we going to have full (level-5) self-driving cars?

What are the most attractive opportunities now in warehouse robotics?

You wrote a famous article in 2017, “The Seven Deadly Sins of AI Prediction.“ You said then that you wanted an artificial general intelligence to exist—in fact, you said it had always been your personal motivation for working in robotics and AI. But you also said that AGI research wasn’t doing very well at that time at solving the basic problems that had remained intractable for 50 years. My impression now is that you do not think the emergence of GPT-4 and other large language models means that an AGI will be possible within a decade or so.

Rodney Brooks: You’re exactly right. And by the way, GPT-3.5 guessed right—I asked it about me, and it said I was a skeptic about it. But that doesn’t make it an AGI.

The large language models are a little surprising. I’ll give you that. And I think what they say, interestingly, is how much of our language is very much rote, R-O-T-E, rather than generated directly, because it can be collapsed down to this set of parameters. But in that “Seven Deadly Sins” article, I said that one of the deadly sins was how we humans mistake performance for competence.

If I can just expand on that a little. When we see a person with some level performance at some intellectual thing, like describing what’s in a picture, for instance, from that performance, we can generalize about their competence in the area they’re talking about. And we’re really good at that. Evolutionarily, it’s something that we ought to be able to do. We see a person do something, and we know what else they can do, and we can make a judgement quickly. But our models for generalizing from a performance to a competence don’t apply to AI systems.

The example I used at the time was, I think it was a Google program labeling an image of people playing Frisbee in the park. And if a person says, “Oh, that’s a person playing Frisbee in the park,” you would assume you could ask him a question, like, “Can you eat a Frisbee?” And they would know, of course not; it’s made of plastic. You’d just expect they’d have that competence. That they would know the answer to the question, “Can you play Frisbee in a snowstorm? Or, how far can a person throw a Frisbee? Can they throw it 10 miles? Can they only throw it 10 centimeters?” You’d expect all that competence from that one piece of performance: a person saying, “That’s a picture of people playing Frisbee in the park.” .... ' 

ChatGPT Plugin Connects to Blockchain

 Not expected ......  Actually gave a talk that mentioned both last month.  

Solana Labs integrates ChatGPT plugin to connect blockchain to AI    BY KYT DOTSON

Solana Labs, the developer of the Solana blockchain, announced Tuesday that it has released a ChatGPT plugin that connects its blockchain technology with artificial intelligence that will allow users to query information from the network.

ChatGPT is an artificial intelligence chatbot from OpenAI LP that can answer questions from users in conversational language. The company made it possible for developers to extend it with plugins in March, which makes it possible for the bot to pull data from external sources.

With this announcement, Solana has become the first Layer 1 blockchain to integrate AI capabilities. Using the plugin, users will be able to query information directly from the blockchain using conversational prompts in order to understand Solana data and protocols. That’s aimed at enabling developers and amateurs to get a better understanding of information about the computing infrastructure, underlying data and decentralized finance projects.

“Every developer building consumer-oriented apps should be thinking about how their app is going to be interacted with through an AI model because this is a new paradigm for telling computers what to do,” said Anatoly Yakovenko, co-founder and chief executive of Solana Labs, said in an interview published on Solana’s website.

Yakovenko added that by using AI, people can almost feel like they’re talking to another human, which can completely change how they digest information. “AI will make Solana more usable and understandable,” he said

The plugin was initially teased by Solana in late April, noting that it would be an open-source implementation allowing users to manipulate data on the blockchain. It currently connects directly to a Solana blockchain node and users can buy nonfungible tokens, transfer tokens, examine transactions on the chain, examine account data and look at NFT prices.

Nonfungible tokens are a type of crypto asset built on blockchain technology that provides provable ownership of digital items, such as artwork and video game items. They are cryptographic tokens that can be bought, sold and traded. NFTs are often part of collections, which people often buy, sell and trade similar to collectible trading cards.

Right now, the AI can describe objects on-chain, such as if it’s part of an NFT set, but it won’t yet automate the creation of NFT collections. Yakovenko said a future goal is to add the ability to write to the blockchain, but the plugin is not there yet. Developers interested in taking the plugin for a spin can grab its current iteration from the GitHub repository.

One of Solana’s claims to value in the blockchain space is that it is a high-speed, low-fee blockchain that makes it easy to deploy and use NFTs on. With the addition of an AI interface for users to talk to and understand the network, Yakovenko hopes, it will attract more developers.   .... ' 

Alien Minds, Immaculate Bullshit, Outstanding Questions: College in the Age of ChatGPT

 From the Penn Alumni Mag,   Where I was much interested in AI. Considerable piece I am reading.  

 Considerble pieice, I do NOT agree in much of it, below an intro the whole thing follows 

Alien Minds, Immaculate Bullshit, Outstanding Questions

26 Apr 2023   Page 22-33

College in the age of ChatGPT.

By Trey Popp | Illustration by Chris Gash   Sidebar | The Coming Economic and Ethical Earthquake

In June 2021, Chris Callison-Burch typed his first query into GPT-3, a natural language processing platform developed by the San Francisco-based company OpenAI. Callison-Burch, an associate professor of computer and information science, was hardly new to AI chatbots or the neural networks that power them. He’s been at the forefront of machine translation since the early 2000s, and at Penn he teaches courses in computational linguistics and artificial intelligence. Besides, digital assistants like Siri and Alexa had already woven NLPs into the fabric of everyday life. But the jaw-dropping fluency of OpenAI’s new model pitched him into a “career existential crisis.”

It could respond to prompts with cogent, grammatically impeccable prose. It could turn plain language into Python code. It could expand bullet-point outlines into five-paragraph essays—or theatrical dialogues.

“I was like, ‘Is there anything left for me to do? Should I just drop out of computer science and become a poet?’” he later recollected. “But then I trained the model to write better poetry than me

On November 30, 2022, OpenAI publicly released a refined version called ChatGPT. Its shock-and-awe debut quickly gave Callison-Burch plenty of company on campus. On February 1 he went to a meeting convened by Penn’s Center for Teaching & Learning (CTL) to address the anxiety and excitement racing through faculty lounges—especially after the bot had passed a Wharton operations management exam administered to it by Christian Terwiesch, the Andrew M. Heller Professor. “It was probably the best-attended CTL meeting ever,” Callison-Burch recalled, with a wry chuckle, a couple weeks later. So many people came that CTL director Bruce Lenthall split them into three sessions—two comprising social sciences and humanities faculty and one that blended professors of math, engineering, and physical sciences with counterparts from the University’s health schools.

They’d come for varied reasons. “Some people were just alarmed,” Lenthall said. Having ingested vast swathes of internet text, ChatGPT and other so-called generative AI tools are exquisitely adapted to serve as “sophisticated plagiarism machines,” in the words of Eric Orts, the Guardsmark Professor in Wharton’s department of legal studies and business ethics, who’d experimented with ChatGPT in an MBA course and discussed it within the Faculty Senate executive committee. Other attendees had yet to engage with the tools at all and simply wanted to learn about them. A third group sensed a chance to get in on the ground floor of a revolutionary change. “They suggested that this could be exciting and open up possibilities,” Lenthall recalled, “but they didn’t really have a good idea of what those might be.”

Most participants fell somewhere in the middle—worried about the threats ChatGPT posed to established modes of teaching and evaluation, but curious about its potential to advance the scope or pace of instruction. “What was the most gratifying to me,” said Lenthall, who is also an adjunct associate professor of history, “was that all the faculty came to the conclusion that they really needed to think through the question: What is it most critical that my students learn to do on their own? And when is it most appropriate for them to do something with another tool?”

Their search for answers gave the spring semester a hothouse atmosphere of probing and experimentation.

Wharton associate professor Ethan Mollick, a Ralph J. Roberts Distinguished Faculty Scholar and academic director of Wharton Interactive, not only permitted but in some cases required students in his innovation and entrepreneurship courses to use generative AI platforms, which he likened to “analytic engines.” Meanwhile, on the other end of campus, astronomy professor Masao Sako and his analytical mechanics students asked ChatGPT to solve homework problems. “It returns answers and explanations that sound plausible,” Sako said, but “failed on every single one.” Given the confident authority with which ChatGPT announced its defective solutions, Sako concluded that the tool might indeed have some utility in the realm of upper-level physics. “I’ve told my students to continue using it to get some practice on identifying errors, which is a useful skill.”

Penn Integrates Knowledge (PIK) University Professor Konrad Kording, who teaches psychology and neuroscience with a focus on neural networks and machine learning, emerged as a pithy generative AI maximalist. “It’s just a mistaken opportunity for any student to not use ChatGPT for any possible project they’re working on,” he declared at a late-February panel discussion sponsored by the School of Arts & Sciences’ Data Driven Discovery Initiative (DDDI). “It’s just incompetent. We should give them bad grades for not using ChatGPT.” Yet Eric Orts was finding that when he let his MBA students use it for some assignments, it tended to lead them toward bad grades—in the form of “deadening” prose—all on its own. “I’m convinced that there are positive uses emerging for this in the real world,” he told me. “But in general I was not impressed by the answers I got from students using it.”

Neither was Karen Rile C’80, a fiction writing teacher in the English department whose experimentation with chatbots goes back to a primitive model developed by AOL at the turn of the century. “What I value in writing is specificity, sharpness, clarity—and it fails on every level,” she reflected. “It’s like a bad student writer who writes in a way that’s very generic, with lots of vague cliches and phrases. It feels blurry. I think that it’ll probably get sharper and better, but I can’t imagine it’s ever going to do anything that’s literary quality. It’ll be very formulaic.”

Yet Rile kicked off her fiction seminar this spring by assigning her students a piece by a writer who’d used GPT-3 as a kind of a coauthor. “I wanted to get ahead of it at the beginning of the semester,” she told me. Then, in mid-March, she brought in Callison-Burch and his PhD student Liam Dugan EAS’20 GEng’20, who focuses on natural language processing, to give a guest lecture about generative AI and creative writing.

All 10 professors I interviewed, plus another four who participated in that DDDI panel discussion and several with whom I spoke informally, expressed a similarly open attitude. Sako’s dim view of ChatGPT’s analytical chops didn’t keep him from seeing its potential to boost the conceptual sophistication of his mid-level coding class. Mollick mixed breathless boosterism with a running list of warnings about its boundless propensity to deceive users. Skeptics were on the lookout for positive use cases, and enthusiasts frequently offered insights about generative AI’s limitations.   .... ' '

Reading a Novel very Quickly.

Caught myself quickly reading a few summarized books with Chat yesterday, ... 

ChatGPT rival Anthropic’s Claude can now read a novel in under a minute  BY KYT DOTSON

OpenAI LP rival chatbot maker Anthropic said Thursday that it’s upgrading its own generative artificial intelligence with a much longer memory to take on the internet’s darlings ChatGPT and Google LLC’s Bard.

Since its debut late last year, generative AI has caught the attention of users and businesses for its ability to hold humanlike conversations and pull insightful information from documents. 

Although these AIs are trained on vast amounts of data, one issue is that conversations with them are limited by the amount of memory that they can handle at one time, known as a “context window.” It determines how much text an AI can handle before it begins to “forget” what was being discussed earlier. This is the same reason that certain AI chatbots, such as Microsoft Corp.’s Bing Chat, have a limited number of responses built into them.

These AIs generally don’t operate in terms of words, and instead use tokens, which break down the text sent to them into parts of words and phrases. Claude has been upgraded from 9,000 tokens to 100,000 tokens, which corresponds to about 75,000 words.

“This means businesses can now submit hundreds of pages of materials for Claude to digest and analyze, and conversations with Claude can go on for hours or even days,” Anthropic’s team said in the announcement.

In comparison, ChatGPT’s context window is approximately 4,000 tokens (or about 3,000 words) and OpenAI LP’s GPT-4 has a context window of 32,768 tokens (about 25,000 words).

To test the greatly enhanced memory for Claude, the Anthropic team fed the entire text of “The Great Gatsby” into Calude-Instant, a low-latency version of the chatbot. But the developers modified one line to see if the AI would catch it. The AI read the entire novel and spotted the difference in under 22 seconds.

The novel is becoming a favorite testing ground for engineers building AI models. MosaicML Inc., a generative AI infrastructure and model provider, also used it to show off the capabilities of its MosaicML foundational model.

“Beyond just reading long texts, Claude can help retrieve information from documents that help your business run,” the Anthropic team wrote. “You can drop multiple documents or even a book into the prompt and then ask Claude questions that require synthesis of knowledge across many parts of the text.”

It’s possible to work with smaller context windows and have an AI digest and provide insights for large documents. It means breaking them into smaller portions and having the AI summarize them so that the summaries themselves can be stitched together into the smaller context window. The problem with doing this is that summarization necessarily leads to picking and choosing what to discard in a specific section of text and its context is then lost for comparison with later sections. That’s not necessary if the entire text can be held in memory at once.  ... ' 

Privacy Problem on the iPhone with ChatGPT

 Been working for a few days now with ChatGPT on the iPhone, nicely done, but then came this. warning for now,  lets fix it!

The ChatGPT iPhone App from OpenAI Has a Glaring Privacy Problem: The Company Can Read Your ConversationsThe app warns you not to send personal information in your prompts.

BY JASON ATEN, TECH COLUMNIST

On Thursday, OpenAI released an iOS app for ChatGPT and it quickly became the most popular free app in the App Store. That's not surprising considering some reports suggest ChatGPT had more than 100 million users in January--just two months after it launched. That would make it the fastest-growing technology product of all time. 

For comparison, it took Facebook four and a half years to reach that number. Even TikTok took nine months. 

Until now, almost all of the usage of ChatGPT has been done in a browser on a laptop or desktop computer. There was no mobile app available, and accessing the website on your iPhone wasn't ideal.

At the same time, there have been plenty of imposter apps attempting to capitalize on the fact that so many people are paying attention to generative A.I. and exploring what it can do. It makes sense that OpenAI would want to get its own app out in the world. 

The official app comes with a few cool features. First, there's the fact that it's free (there's a paid upgrade to ChatGPT Plus which gets you access to OpenAI's latest language model). Considering that many of the existing apps charge a weekly subscription fee--making them very expensive if not outright scams--having an official app that doesn't cost anything is a welcome development.

The other feature is that you can talk to ChatGPT. ChatGPT can't actually process audio prompts, so the feature will convert your speech to text and send it like any other question. For a conversational A.I. product, being able to simply say what you want to ask is a great feature.

The iOS app does, however, come with one important tradeoff that users should be aware of. It's a big enough deal that the app prompts you the first time you open it. In addition to a caution that ChatGPT may just make things up, there's another warning about sharing personal information because "Anonymized chats may be reviewed by our Al trainers to improve our systems."  .... '