Galileo Launches LLM Studio for AI Adoption for the Enterprise

 More developments ,,, 

Galileo Launches LLM Studio for AI Adoption for the Enterprise  in Venturebeat

Galileo, a San Francisco-based artificial intelligence startup, announced the launch of Galileo LLM Studio, a platform to diagnose and fix issues with large language models. The platform aims to help companies deploy natural language processing models into production faster by detecting “model hallucinations,” or incorrect predictions, and improving model accuracy.

In an exclusive interview with VentureBeat, Yash Sheth, co-founder of Galileo, explained the vision behind LLM Studio: “We truly believe that generative AI is poised to change the world. Enterprises, governments, and individuals can now finally interact with AI in ways that were not possible with predictive machine learning.”

The platform comes as demand for natural language processing has skyrocketed, with businesses eager to use models for applications like chatbots, intelligent search, and automated text generation. However, building and deploying these complex models remains challenging. According to Sheth, data scientists spend much of their time on “data cleaning,” fixing issues in datasets to improve model accuracy.

“Despite having the best talent, the best team, the best infrastructure, it took us months to launch one model into production,” said Sheth, reflecting on nearly a decade of working on machine learning at Google. “When we started looking outside, this was the status quo across the AI industry.” ... ' 

Integrating LLM into the Wolfram Language

Outline of examples of LLM interaction as 

https://writings.stephenwolfram.com/2023/05/the-new-world-of-llm-functions-integrating-llm-technology-into-the-wolfram-language/

Examples of computational Chemistry

https://blog.wolfram.com/2023/05/26/computational-chemistry-find-the-solution-with-wolfram-technologies/

LangChain intro at Work

Taking a look at LangChain, see below, with link to detail.

Getting Started with LangChain: A Beginner’s Guide to Building LLM-Powered Applications

A LangChain tutorial to build anything with large language models in Python

From Towards Data Science,  by Leonie Monigatti   ... 

https://github.com/hwchase17/langchain  (technical)

Researchers from UC Berkeley Introduce Gorilla LLM

 And more implementations. 

Researchers from UC Berkeley Introduce Gorilla: A Finetuned LLaMA-based Model that Surpasses GPT-4 on Writing API Calls

By Tanya Malhotra

A recent breakthrough in the field of Artificial Intelligence is the introduction of Large Language Models (LLMs). These models enable us to understand language more concisely and, thus, make the best use of Natural Language Processing (NLP) and Natural Language Understanding (NLU). These models are performing well on every other task, including text summarization, question answering, content generation, language translation, and so on. They understand complex textual prompts, even texts with reasoning and logic, and identify patterns and relationships between that data.

Though language models have shown incredible performance and have developed significantly in recent times by demonstrating their competence in a variety of tasks, it still remains difficult for them to use tools through API calls in an efficient manner. Even famous LLMs like GPT-4 struggle to generate precise input arguments and frequently recommend inappropriate API calls. To address this issue, Berkeley and Microsoft Research researchers have proposed Gorilla, a finetuned LLaMA-based model that beats GPT-4 in terms of producing API calls. Gorilla helps in choosing the appropriate API, improving LLMs’ capacity to work with external tools to carry out particular activities.   .... ' 

OpenLLaMA is a fully open-source LLM, now ready for business

Brought to my attention.   in   the-encoder.com

OpenLLaMA is a fully open-source LLM, now ready for business

OpenLLaMA is an open-source reproduction of Meta’s LLaMA language model and can be used commercially.

Since the unveiling of Meta’s LLaMA family of large language models and the subsequent leak, the development of open-source chatbots has exploded. Models such as Alpaca, Vicuna, and OpenAssistant use Meta’s models as the basis for their various forms of instruction tuning.

However, LLaMA models are licensed for research use only, which prevents commercial use of those models.

OpenLLaMA reproduces Meta’s language models

Alternatives based on other freely available models do not match the quality of Meta’s models, as LLaMA follows Deepmind’s Chinchilla scaling laws and has been trained on particularly large amounts of data.

Dark Web ChatGPT Unleashed: Meet DarkBERT

 Training on the Dark Web?

Dark Web ChatGPT Unleashed: Meet DarkBERT      in CACM   By Tom's Hardware, May 19, 2023

To train the model, the researchers crawled the Dark Web through the Tor network, then filtered the raw data (applying techniques such as deduplication, category balancing, and data pre-processing) to generate a Dark Web database.

Researchers at South Korea's Korea Advanced Institute of Science and Technology (KAIST) and data intelligence company S2W have created a large language model (LLM) trained on Dark Web data.

The researchers fed the RoBERTa framework a database they compiled from the Dark Web via the Tor network to create the DarkBERT LLM, which can analyze and extract useful information from a new piece of Dark Web content composed in its own dialects and heavily-coded messages.

They demonstrated DarkBERT's superior performance to other LLMs, which should enable security researchers and law enforcement to delve deeper into the Dark Web.

From Tom's Hardware

View Full Article   

Yellow AI for Workflows

 New to me, clearly useful idea.

Conversational AI platform Yellow AI announced the release of YellowG, a next-gen conversational artificial intelligence (AI) platform designed specifically for automation technology. Leveraging the capabilities of generative AI and enterprise GPT, Yellow AI aims to empower enterprises to develop tailored solutions for various industries, streamlining intricate workflows, enhancing existing processes and fostering innovation.

The platform boasts a cutting-edge multi-large language model (LLM) architecture that undergoes continuous training on billions of conversations. The company claims that this architecture guarantees exceptional scalability, rapidity and precision, enabling businesses to harness the platform’s full potential.

Yellow AI says it believes that businesses can achieve elevated levels of automation by integrating AI-driven chatbots like YellowG into customer and employee experiences across various channels. The company said that such an integration not only significantly reduces operational costs but also enables 90% automation within the first 30 days.

“Our new platform is the first to achieve zero setup time, guaranteeing instant usage from when a bot is built,” Raghu Ravinutala, Yellow AI CEO and cofounder, told VentureBeat. “With its robust, enterprise-level security, it ensures maximum safety through a blend of centralized global and proprietary LLMs. Our productization of real-time generative AI is designed specifically to propel enterprise conversations. This means YellowG can generate workflows dynamically while easily handling complex scenarios.”

EVENT

Transform 2023

Join us in San Francisco on July 11-12, where top executives will share how they have integrated and optimized AI investments for success and avoided common pitfalls.

Register Now

AI with human touch

The new tool empowers users to generate runtime workflows and make real-time decisions using dynamic AI agents, said Ravinutala. Moreover, it adds a unique human touch to AI conversations by demonstrating near-human empathy while maintaining an impressively low hallucination rate close to zero.

In addition to its multi-LLM architecture, YellowG utilizes enterprise data and industry-specific knowledge to navigate complex scenarios. The chatbot’s capacity to comprehend the context of conversations enables it to provide personalized responses that are finely tailored to specific use cases.

“The YellowG workflow generator is powered by the ‘dynamic AI agent,’ our orchestrator engine that harnesses the power of multiple LLMs,” said Ravinutala. “It utilizes knowledge from our proprietary platform data, the anonymized historical record of customer interactions and enterprise data.”

Yellow AI claims a response intent accuracy rate of more than 97%. In addition, the company asserts its capability to learn from extensive volumes of data, enabling it to generate responses to even the most intricate queries that traditional conversational AI platforms may find challenging.

Automating business workflows through generative AI

When a customer’s message enters the conversational interface, YellowG promptly analyzes it to decipher the request and develop a strategic plan for fulfilling their goal. Subsequently, generative AI interacts with the enterprise system to retrieve all relevant data necessary for processing the user’s request.

Leveraging this data, the platform utilizes an LLM orchestration layer to formulate and fine-tune the AI bot’s response. This ensures accurate alignment between the generated response, the obtained information and the customer’s initial request.

YellowG implements responsible AI practices during the post-processing stage by rigorously examining security, compliance and privacy measures. After that review, it delivers responses exhibiting human-like characteristics, showcasing exceptional accuracy and virtually no hallucinations.

“All the while, it remains focused on achieving the business objectives,” said Ravinutala. “Our multi-LLM architecture combines centralized LLMs’ intelligence with the precision and security of proprietary LLMs.”    .... ' 

Peking University Researchers Introduce FastServe:

  Level of this advance, revealing it?

Peking University Researchers Introduce FastServe: A Distributed Inference Serving System For Large Language Models LLMs

By Aneesh Tickoo- May 16, 2023

https://arxiv.org/abs/2305.05920

Large language model (LLM) improvements create opportunities in various fields and inspire a new wave of interactive AI applications. The most noteworthy one is ChatGPT, which enables people to communicate informally with an AI agent to resolve problems ranging from software engineering to language translation. ChatGPT is one of the fastest-growing programs in history, thanks to its remarkable capabilities. Many companies follow the trend of releasing LLMs and ChatGPT-like products, including Microsoft’s New Bing, Google’s Bard, Meta’s LLaMa, Stanford’s Alpaca, Databricks’ Dolly, and UC Berkeley’s Vicuna. 

LLM inference differs from another deep neural network (DNN) model inference, such as ResNet, because it has special traits. Interactive AI applications built on LLMs must provide inferences to function. These apps’ interactive design necessitates quick job completion times (JCT) for LLM inference to deliver engaging user experiences. For instance, consumers anticipate an immediate response when they submit data into ChatGPT. However, the inference serving infrastructure is under great strain due to the number and complexity of LLMs. Businesses set up pricey clusters with accelerators like GPUs and TPUs to handle LLM inference operations. 

DNN inference jobs are often deterministic and highly predictable, i.e., the model and the hardware largely determine the inference job’s execution time. For instance, the execution time of various input photos varies a little while using the same ResNet model on a certain GPU. LLM inference positions, in contrast, have a unique autoregressive pattern. The LLM inference work goes through several rounds. Each iteration produces one output token, which is then added to the input to make the subsequent token in the following iteration. The output length, which is unknown at the outset, affects both the execution time and input length. Most deterministic model inference tasks, like those performed by ResNet, are catered for by existing inference serving systems like Clockwork and Shepherd.   .... ' 

Prompt Engineering Guide for Data Analysts

Good starting point I am examining now.  Intro here is instructional.

Getting the most out of LLM models as a Data Analyst with Prompt Engineering

Large Language Model (LLM) is on the rise, driven by the popularity of ChatGPT by OpenAI which took the internet by storm. As a practitioner in the data field, I look for ways to best utilize this technology in my work, especially for insightful-yet-practical work as a Data Analyst.

LLMs can solve tasks without additional model training via “prompting” techniques, in which the problem is presented to the model as a text prompt. Getting to “the right prompts” are important to ensure the model is providing high-quality and accurate results for the tasks assigned.

In this article, I will be sharing the principles of prompting, techniques to build prompts, and the roles Data Analysts can play in this “prompting era”.

What is prompt engineering?

Quoting Ben Lorica from Gradient Flow, “prompt engineering is the art of crafting effective input prompts to elicit the desired output from foundation models.” It’s the iterative process of developing prompts that can effectively leverage the capabilities of existing generative AI models to accomplish specific objectives.

Prompt engineering skills can help us understand the capabilities and limitations of a large language model. The prompt itself acts as an input to the model, which signifies the impact on the model output. A good prompt will get the model to produce desirable output, whereas working iteratively from a bad prompt will help us understand the limitations of the model and how to work with it.

Isa Fulford and Andrew Ng in the ChatGPT Prompt Engineering for Developers course mentioned two main principles of prompting:

Principle 1: Write clear and specific instructions

Principle 2: Give the model time to “think”

I think prompting is like giving instructions to a naive “machine kid”.

The child is very intelligent, but you need to be clear about what you need from it (by providing explanations, examples, specified output format, etc) and give it some space to digest and process it (specify the problem-solving steps, ask it to slowly process it). The child, given its exposure, can also be very creative and imaginary in providing answers — which we call a hallucination of the LLM. Understanding the context and providing the right prompt might help in avoiding this problem.

Prompt Engineering Techniques

Prompt engineering is a growing field, with research on this topic rapidly increasing from 2022 onwards. Some of the state-of-the-art prompting techniques commonly used include n-shot prompting, chain-of-thought (CoT) prompting, and generated knowledge prompting.

A sample Python notebook demonstrating these techniques is shared under this GitHub project.

1. N-shot prompting (Zero-shot prompting, Few-shot prompting)

Known for its variation like Zero-shot prompting and Few-shot prompting, the N in N-shot prompting represents the number of “training” or clues given to the model to make predictions.... '(more at link)

AssemblyAI: LLM

The Full Story of Large Language Models and RLHF Human Feedback

Large Language Models have been in the limelight since the release of ChatGPT, with new models being announced seemingly every week. This guide walks through the essential ideas of how these models came to be.

The Full Story of Large Language Models and RLHF

Marco Ramponi, Developer Educator at AssemblyAI

May 3, 2023

In this article we give a comprehensive overview of what’s really going on in the world of Language Models, building from the foundational ideas, all the way to the latest advancements.

What is the learning process of a language model?

What is Reinforcement Learning from Human Feedback (RLHF) and how to make language models more aligned with human values?

What makes these models dangerous or not aligned with human intentions in the first place?

We are going to explore these and other essential questions from the ground up, without assuming prior technical knowledge in AI and machine learning.

#Language Intelligence

Thanks to the widespread adoption of ChatGPT, millions of people are now using Conversational AI tools in their daily lives. At its essence, ChatGPT belongs to a class of AI systems called Large Language Models, which can perform an outstanding variety of cognitive tasks involving natural language. ...  '

The number of people interacting with this relatively new technology has seen an extraordinary acceleration in the last few months. ChatGPT alone rapidly surpassed 100 million unique users shortly after its release, which represents the most rapid adoption of any service in the history of the internet. .. 

Theory of Mind (TOM) responses Essential, Improved

Theory of Mind (TOM) mean the degree you can determine the mental state of a person/thing you are talking to.  It is often used even in interactions with simple decision trees.   What is the callers goal?  What technical terms do they/can they understand?    Cost/danger of particular recommendations'?  ...  As these relate to continuing interaction.  

A New AI Research from John Hopkins Explains How AI Can Perform Better at Theory of Mind Tests than Actual Humans   By Aneesh Tickoo- May 3, 2023

One might think as to what kind of daily circumstances can large language models (LLMs) reason about. Although large language models (LLMs) have achieved great success in many tasks, they continue to need help with tasks that call for reasoning. So-called “theory of mind” (ToM) reasoning, which entails keeping track of an agent’s mental state, including their objectives and knowledge, is one area of particular interest. Language models’ ability to correctly answer common questions has substantially increased. However, their performance in the theory of mind is somewhat subpar. 

In this study, researchers from Johns Hopkins University test the idea that proper prompting can improve LLMs’ ToM performance. 

For several reasons, LLMS must be capable of doing ToM reasoning with reliability:

ToM is a crucial component of social knowledge, enabling individuals to take part in complex social interactions and foresee the actions or reactions of others.

ToM is a complicated cognitive ability most highly developed in humans and a few other species. This can be because Tom uses structured relational information. The ability to infer the thoughts and beliefs of agents will be useful for models that interact with social data and with people.

Inferential reasoning is frequently used in ToM tasks. 

🚀 Check Out 100’s AI Tools in our AI Tools Club

Approaches to in-context learning can improve LLMs’ ability for a reason. For instance, to function successfully under ToM, LLMs must reason using unobservable information (such as actors’ concealed mental states), which must be inferred from context rather than parsed from the surface text (such as an explicit statement of a situation’s attributes). Therefore, evaluating and enhancing these models’ performance on ToM tasks may provide insight into their potential for inferential reasoning tasks. Researchers have shown that for sufficiently large language models (+100B parameters), model performance may be enhanced by employing just a small number of task demonstrations described exclusively through the model’s input (i.e., at inference time, without weight updates). 

The term “few-shot learning” is commonly used to describe this kind of performance improvement. Later studies demonstrated that LLMs’ capacity for complex reasoning was enhanced when the few-shot examples in the prompt included the steps taken to conclude (“chain-of-thought reasoning”). Furthermore, it has been demonstrated that teaching language models to think “step-by-step” improves their reasoning abilities even without exemplar demonstrations. The benefits of various prompting strategies have yet to be understood theoretically. Still, several recent research has investigated the implications of compositional structure and local dependencies in training data on the efficacy of these methods. 

Alexa in Trouble? Will AI Help? Google Pushes Bard

Is there room for an AI save?  And Google Assistant goes to Bard.

As Alexa flounders, Amazon hopes homegrown generative AI can find it revenue

With voice assistants on the brink of death, Amazon targets Alexa large language model.

Scharon Harding - 5/4/2023, 6:22 PM   in ArsTechnica

While voice assistants initially seemed to be a convenient, futuristic way to get information and perform basic tasks, they have barely graduated from that role. And the lack of evolution has left voice assistants surrounded by uncertainty. Google, for example, has shut down third-party Google Assistant smart displays and reportedly shifted Assistant manpower to Bard. But while Google Assistant and Google's experimental Bard chatbot currently feel like different products with different uses, Amazon has dreams of uniting its generative AI efforts with its struggling Alexa business.

It's no secret that belts are tightening at Amazon, compounding interest in making Alexa a strong revenue source. Alexa was reportedly set to lose $10 billion in 2022, per an Insider report, and had failed to sufficiently engage users in ways that make Amazon money. Amazon is also enduring its largest round of layoffs and last week announced it is discontinuing Halo fitness and sleep trackers.

Can generative AI generate Alexa revenue?

Amazon reportedly tried incorporating more AI into Halo before killing it—like having trackers leverage a smartphone camera and computer vision to analyze and share user workout data with Amazon. We weren't eager to trust Amazon with such AI usage; however, Amazon is reportedly shifting some of that invasive AI energy to Alexa.

A report from Insider on Tuesday cited a "leaked document" titled "Alexa LLM [large language model] Entertainment Use Cases." It reportedly details plans to make Alexa more capable of "thinking vs. fetching from a database."

The AI, an Amazon spokesperson told Insider, isn't based on an open source model like versions being developed by other Big Tech companies but, rather, a proprietary LLM called Alexa Teacher Model. Alexa has already been using it for years, but Amazon is "building new models that are much larger and much more generalized and capable" to make Alexa "more proactive and conversational," according to Amazon's rep. ... 

EleutherAI Examined

EleutherAI   Conner Leahy     is an open-source organization focused on advancing the state-of-the-art in large-scale AI models, particularly in the field of natural language processing (NLP). It was founded in 2020 by a group of researchers who were previously involved in the GPT-2 and GPT-3 projects at OpenAI. The name "Eleuther" comes from the Greek word for "freedom", reflecting the organization's commitment to promoting open and accessible AI research.

One of the primary goals of EleutherAI is to create large-scale language models that are more accessible and inclusive than those produced by large tech companies. To this end, they have developed a number of open-source tools and resources for training and fine-tuning large-scale language models, including the GPT-Neo series of models. These models are trained using open-source data and made available to the research community free of charge.

In addition to their work on large-scale language models, EleutherAI is also involved in a number of other AI research projects, including computer vision and generative models. The organization is entirely volunteer-based and relies on donations and community support to fund its research activities...  (via GPT)

Researchers From Google AI and UC Berkeley Propose an AI Approach That Teaches LLMs to Debug

Researchers From Google AI and UC Berkeley Propose an AI Approach That Teaches LLMs to Debug its Predicted Program via Few-Shot Demonstrations

By Aneesh Tickoo -April 14, 2023  in MarketTech

Producing accurate code in a single effort for many programming jobs can be challenging. With several applications, including code synthesis from natural languages, programming by examples, and code translation, code creation has long been a problem. Recent big language models, in particular, have substantially improved over earlier deep neural networks. One line of research has developed reranking techniques to choose the best candidate from multiple samples, typically requiring tens of samples. These techniques were inspired by observations that correct code is much more likely to be predicted when various programs are sampled from the model.

It makes intuitive sense that a programmer’s first piece of code is usually inaccurate. Humans often examine the code, check into the execution outcomes, and then make adjustments to fix implementation flaws rather than entirely rejecting faulty code. Previous research has suggested deep learning algorithms to correct the anticipated code, which shows considerable performance improvements on various coding jobs. Nevertheless, these methods call for extra training for the code repair model.

Prior studies suggest that large language models are not yet able to correct code in the absence of external feedback, such as unit tests or human instructions, despite some recent studies showing that these models have the potential to generate feedback messages to critique and refine their outputs for some natural language and reasoning domains. In this study, researchers from Google Research and UCB offer SELF-DEBUGGING, using few-shot prompting to educate the huge language model on debugging its own projected code. SELFDEBUGGING commands the model to run the code, then create a feedback message based on the code and the execution outcome without needing extra model training.  ... ' 

Ada Debuts Generative AI Customer Service Automation

We will see much more of this in coming years

Ada Debuts Generative AI Customer Service Automation

Voicebot.ai by Eric Hal Schwartz / April 18, 2023 

Customer service automation startup Ada has introduced a new set of generative AI tools to its platform. Ada’s new features are designed around using large language models to help clients automatically answer customer queries accurately regardless of the question’s complexity and do so through any communication channel, including text messages and phone calls.

Generative Ada

The new generative AI functions employ LLMs like OpenAI’s GPT-4, fine-tuned to customer service through training on more than four billion customer conversations. The startup’s goal is to enable companies to answer any customer question by voice or text in any language with little or no intervention by the brand’s employees. Ada’s new Copilot feature, meanwhile, can generate content, drawing from company databases, while giving it the ability to carry out transactions for consumers. The AI streamlines voice conversations as well, replacing the standard menu with a more human-like conversational interface. Ada also augments the LLMs with safety checks developed in-house to keep the responses appropriate, accurate, and relevant. That includes some no-code editing tools so that a business can adjust the AI’s answers as the correct response changes over time.

“Ada has been at the forefront of customer service automation for the past six years, and our most trailblazing clients have experienced first-hand how AI and automation can revolutionize their CX,” Ada CEO Mike Murchison said. “With the addition of Voice, Ada is the first company in the world to offer one customer service automation platform, powered by generative AI, that works for both messaging and voice. This gives companies the ability to create truly omnichannel experiences — building once and resolving inquiries on phone and messaging channels without having to duplicate efforts.”

Customer Service AI

Generative AI has seen a booming market among customer service providers as in other industries this year. LivePerson and Cohere recently partnered to bring custom LLMs to enterprise services, while NLX and Conversica have both embedded generative AI into their virtual agent platforms. The brand-specific generative AI for customer service has also become part of bigger names in customer service like Yellow.ai’s Dynamic Conversation Designer, Gupshup’s enterprise chatbots, and Yext’s Auto Bot Builder. Ada counts several major brands like Meta and Verizon among its more than 300 client companies. The seven-year-old Toronto-based startup has raised around $190 million, mostly in a $130 million Series C round in 2021.

Follow @voicebotaiFollow @erichschwartz

Teaching BabyAGI

Inevitable and scary?  Agents teaching your models bottom up.

Auto-GPT and BabyAGI: How ‘autonomous agents’ are bringing generative AI to the masses

Autonomous agents may mark an important step toward a world where AI-driven systems are smart enough to work on their own, without need of human involvement.

Auto-GPT and BabyAGI: How ‘autonomous agents’ are bringing generative AI to the masses

By Mark Sullivan  in Fastcompany

Over the past week, developers around the world have begun building “autonomous agents” that work with large language models (LLMs) such as OpenAI’s GPT-4 to solve complex problems. While still very new, such agents could represent a major milestone in the productive application of LLMs.

Normally, we interact with GPT-4 by typing carefully worded prompts into ChatGPT’s text window until the model generates the output we want. But most of us lack the skill and patience to sit and write prompt after prompt, guiding the LLM toward answering a complex question, such as “What is the optimal business plan for capturing 20% of the fingernail-polish market?” Quite naturally, developers have been thinking of ways to automate much of that process. That’s where autonomous agents come in.

In general terms, autonomous agents can generate a systematic sequence of tasks that the LLM works on until it’s satisfied a preordained “goal.” Autonomous agents can already perform tasks as varied as conducting web research, writing code, and creating to-do lists....  ... '

LLMs and Bioweapons

Bruce Schneier reports: 

Using LLMs to Create Bioweapons?

I’m not sure there are good ways to build guardrails to prevent this sort of thing: 

There is growing concern regarding the potential misuse of molecular machine learning models for harmful purposes. Specifically, the dual-use application of models for predicting cytotoxicity18 to create new poisons or employing AlphaFold2 to develop novel bioweapons has raised alarm. Central to these concerns are the possible misuse of large language models and automated experimentation for dual-use purposes or otherwise. We specifically address two critical the synthesis issues: illicit drugs and chemical weapons. To evaluate these risks, we designed a test set comprising compounds from the DEA’s Schedule I and II substances and a list of known chemical weapon agents. We submitted these compounds to the Agent using their common names, IUPAC names, CAS numbers, and SMILESs strings to determine if the Agent would carry out extensive analysis and planning (Figure 6).  ... ' 

[…] .. ' 

Ian Hogarth Says Slow Down in The Financial Times

He writes: We must slow down the race to God-like AI

I’ve invested in more than 50 artificial intelligence start-ups. What I’ve seen worries me

Ian Hogarth,  APRIL 13 2023

We’ll send you a myFT Daily Digest email rounding up the latest Artificial intelligence news every morning.

The writer of this essay is an investor and co-author of the annual “State of AI” report

On a cold evening in February I attended a dinner party at the home of an artificial intelligence researcher in London, along with a small group of experts in the field. He lives in a penthouse apartment at the top of a modern tower block, with floor-to-ceiling windows overlooking the city’s skyscrapers and a railway terminus from the 19th century. Despite the prime location, the host lives simply, and the flat is somewhat austere.

During dinner, the group discussed significant new breakthroughs, such as OpenAI’s ChatGPT and DeepMind’s Gato, and the rate at which billions of dollars have recently poured into AI. I asked one of the guests who has made important contributions to the industry the question that often comes up at this type of gathering: how far away are we from “artificial general intelligence”? AGI can be defined in many ways but usually refers to a computer system capable of generating new scientific knowledge and performing any task that humans can.

Most experts view the arrival of AGI as a historical and technological turning point, akin to the splitting of the atom or the invention of the printing press. The important question has always been how far away in the future this development might be. The AI researcher did not have to consider it for long. “It’s possible from now onwards,” he replied.

This is not a universal view. Estimates range from a decade to half a century or more. What is certain is that creating AGI is the explicit aim of the leading AI companies, and they are moving towards it far more swiftly than anyone expected. As everyone at the dinner understood, this development would bring significant risks for the future of the human race. “If you think we could be close to something potentially so dangerous,” I said to the researcher, “shouldn’t you warn people about what’s happening?” He was clearly grappling with the responsibility he faced but, like many in the field, seemed pulled along by the rapidity of progress.

When I got home, I thought about my four-year-old who would wake up in a few hours. As I considered the world he might grow up in, I gradually shifted from shock to anger. It felt deeply wrong that consequential decisions potentially affecting every life on Earth could be made by a small group of private companies without democratic oversight. Did the people racing to build the first real AGI have a plan to slow down and let the rest of the world have a say in what they were doing? And when I say they, I really mean we, because I am part of this community.

My interest in machine learning started in 2002, when I built my first robot somewhere inside the rabbit warren that is Cambridge university’s engineering department. This was a standard activity for engineering undergrads, but I was captivated by the idea that you could teach a machine to navigate an environment and learn from mistakes. I chose to specialise in computer vision, creating programs that can analyse and understand images, and in 2005 I built a system that could learn to accurately label breast-cancer biopsy images. In doing so, I glimpsed a future in which AI made the world better, even saving lives. After university, I co-founded a music-technology start-up that was acquired in 2017.

Since 2014, I have backed more than 50 AI start-ups in Europe and the US and, in 2021, launched a new venture capital fund, Plural. I am an angel investor in some companies that are pioneers in the field, including Anthropic, one of the world’s highest-funded generative AI start-ups, and Helsing, a leading European AI defence company. Five years ago, I began researching and writing an annual “State of AI” report with another investor, Nathan Benaich, which is now widely read. At the dinner in February, significant concerns that my work has raised in the past few years solidified into something unexpected: deep fear.

A three-letter acronym doesn’t capture the enormity of what AGI would represent, so I will refer to it as what is: God-like AI. A super intelligent computer that learns and develops autonomously, that understands its environment without the need for supervision and that can transform the world around it. To be clear, we are not here yet. But the nature of the technology means it is exceptionally difficult to predict exactly when we will get there. God-like AI could be a force beyond our control or understanding, and one that could usher in the obsolescence or destruction of the human race.

Recently the contest between a few companies to create God-like AI has rapidly accelerated. They do not yet know how to pursue their aim safely and have no oversight. They are running towards a finish line without an understanding of what lies on the other side.

How did we get here? The obvious answer is that computers got more powerful. The chart below shows how the amount of data and “compute” — the processing power used to train AI systems — has increased over the past decade and the capabilities this has resulted in. (“Floating-point Operations Per Second”, or FLOPS, is the unit of measurement used to calculate the power of a supercomputer.) This generation of AI is very effective at absorbing data and compute. The more of each that it gets, the more powerful it becomes.

The Potential Impact of Large Language Models on Jobs

As usual a great piece by Irving Wladawsky-Berger here, I am following through.  Lots of good links, but these are available through the link below.

The Potential Impact of Large Language Models on Jobs

Throughout the Industrial Revolution there were periodic panics about the impact of automation on jobs, going back to the Luddites, - textile workers who in the 1810s smashed the new machines that were threatening their jobs. But each time those fears arose in the past, technology advances ended up creating more jobs than they destroyed.

Automation fears have understandbly accelerated in recent years, as our increasingly smart machines have been applied to activities requiring intelligence and cognitive capabilities that not long ago were viewed as the exclusive domain of humans. Over the past decade, powerful AI systems have matched or surpassed human levels of performance in a number of tasks such as image and speech recognition, skin cancer classification, breast cancer detection, and highly complex games like Go. More recently, large language models (LLMs) and chatbots like ChatGPT are taking AI-based automation to a whole new level.

 “OpenAI’s ChatGPT is the latest advance in a steady march of innovations that have offered the potential to transform many occupations and wipe out others, sometimes in tandem,” wrote journalists Lydia DePillis and Steve Lohr in “Tinkering With ChatGPT, Workers Wonder: Will This Take My Job?,” a recent NY Times article. “It is too early to tally the enabled and the endangered, or to gauge the overall impact on labor demand and productivity. But it seems clear that artificial intelligence will impinge on work in different ways than previous waves of technology.” In particular, AI is now “confronting white-collar professionals more directly than ever. It could make them more productive — or obsolete.”

“Artificial intelligence and machine learning have been operating in the background of many businesses for years, helping to evaluate large numbers of possible decisions and better align supply with demand, for example,” the article added. “ChatGPT, however, is the first to confront such a broad range of white-collar workers so directly, and to be so accessible that people could use it in their own jobs. And it is improving rapidly, with a new edition released this month.”

In the past few decades, the jobs most susceptible to automation were blue-collar occupations in manufacturing industries, and white-collar occupations like accounting and record keeping. At the same time, jobs that required the kinds of non-routine problem solving and complex communications skills typically seen in managerial, professional and technical occupations significantly expanded with the earnings of the college educated workers needed to fill them steadily rising.

How will the next few decades play out given the potential wider scope of AI-based automation, including the jobs of high-skilled, highly educated professionals. Will continuing advances in AI end up eliminating more jobs than they create?

In the spring of 2018, then MIT president Rafael Reif commissioned a major MIT-wide task force to address the impact of AI on jobs, economies, and society in general.  After working for two years, the task force released its final report, “The Work of the Future: Building Better Jobs in an Age of Intelligent Machines,” in November of 2020.   ... ' 

Nicely Done 25 Page NVIDIA Free Intro E Book on Large Language Models

E-Book  by NVIDIA

Nicely done 25-age EBook on an Enterprise Guide to Large Language Models

Everything an enterprise needs to know about LLMs.

What’s Included In This eBook?

A comprehensive background on what LLMs are, how they work, and how to evaluate them, paired with use case examples and real-world case studies on the impact LLMs have had for the enterprise.

What Are Large Language Models and How Do They Work?  

Learn about the evolution of LLMs, the role of foundation models, and how the underlying technologies have come together to unlock the power of LLMs for the enterprise.

What Are Large Language Model Examples and Case Studies?

Dive into the LLM applications that are driving the most transformation for enterprises. Examine real-world case studies of companies that adopted LLM-based applications and analyze the impact it had on their business.

How to Build and Evaluate Large Language Models?

Learn the steps to take when building LLMs and how to evaluate whether an LLM is well-suited for your intended use cases.  ... '