Are large language models what we think they are? Some of the top AI researchers in the world choose to stand out from the ChatGPT celebration, claiming it is not worth the hype.  

This article explores what some AI leaders opine about OpenAI's controversial model and other LLMs dominating the AI ecosystem.   

Andrej Karpathy: Andrej Karpathy is the Former Director of AI at Tesla and has recently rejoined OpenAI. He shared a video lecture on "Let's build GPT: from scratch, in code, spelled out." In Karpathy's opinion, when humans generate text, they spend very different amounts of time per token, creating intermediate work, making edits, etc. This is very different from GPTs that go chunk by chunk.  

He expressed his statement in a tweet that the deepest unintuitive disconnect about the psychology of ChatGPT is that it doesn't get "time to think". It has a small, fixed amount of thought for each output token. It resembles a human forced to speak very fast. Asking them to produce more text gives them more time to think. 

Andrew Ng: Andrew Ng, Founder and CEO of Landing AI and the founder of Deep Learning AI, shares his experiment with ChatGPT on Twitter. He states, "ChatGPT is sometimes amazing and sometimes hilariously wrong. 

In Andrew Ng's opinion, there is a path for LLMs to transform how we access information, albeit one that poses technical and business hurdles. 'Large language models like Galactica and ChatGPT can spout nonsense in a confident, authoritative tone. However, this overconfidence - which reflects the data they're trained on - makes them more likely to mislead', he writes. Building large language models that can accurately decide when to be confident and when not will reduce their risk of misinformation and build trust.

Gary Marcus: Gary Marcus, AI researcher, best-selling author, and entrepreneur, writes in his blog that ChatGPT has made all the same kinds of mistakes that its predecessors did. "This was inevitable," points out Marcus. Gary Marcus, along with Ernest Davis, a Professor of Computer Science at New York University, has been working with ChatGPT to study its capabilities and limitations.  

Gary Marcus, in his blog, states that GPT-3 has no idea how the world works. He calls GPT-3, ‘king of pastiche’- meaning, a work of visual art, literature, theatre, music or architecture that imitates the style or character of the work of one or more other artists. 

Kate Crawford: Kate Crawford, professor and author of the book "Atlas of AI", believes that though present large language models are progressive, it also points out genuine problems we are yet to contend with. Speaking about her concerns in a podcast with the Economists, Crawford analyzes the political implications of the model. "Who can afford to build something at this scale?" she asks. 

Models like GPT-3 and Dall-E are built by a tiny handful of the world's largest tech companies. This denotes a concentration of power among a few. Users experience the results these models produce but are blindsided by how they work. Their proprietary nature is a matter of concern when applied to sensitive areas like education, healthcare, and justice.  

Meredith Whittaker: Meredith Whittaker is the President of the Signal Foundation and serves on their Board of Directors. The former Google employee is one of the biggest critics of the big tech. She stated in her paper titled 'The Steep Cost of Capture', that 'in considering how to tackle this onslaught of industrial AI, we must first recognize that the "advances" in AI celebrated over the past decade was not due to fundamental scientific breakthroughs in AI techniques. They were and are primarily the product of significantly concentrated data and compute resources that reside in the hands of a few large tech corporations.'  

In a tweet, Whittaker criticized OpenAI's attempt to release the tool to detect AI-generated text. She wrote, "they created another SOTA benchmark that will be used to measure "improvements" in generative large language models like GPT based on how "undetectable" their outputs are". 

Stewart Russel: Stewart Russel, renowned computer science professor at the University of Berkley, California, states that current models like ChatGPT can perform outside their pre-designated parameters, and the public has been "fooled" into thinking that they can bring about the next evolution of AI. He was speaking at the World Artificial Intelligence Cannes Festival (WAICF).  

Russel opines that while ChatGPT can generate responses to questions, the chatbot "does not know anything". He tagged the model as "remarkably impressive" at generating text but reminded that it could be easily fooled. 

Yann LeCun: Yann LeCun, VP and Chief AI Scientist at Meta, has stated his unwavering opinion on current LLMs. 'ChatGPT is not a step towards human-level-AI', tweeted LeCun. He believes that though they are useful as writing aids, they are "reactive" and don't plan or reason. 'They make stuff up or retrieve stuff approximately, that can be mitigated but not fixed by human feedback', states LeCun in a Facebook post.  

Current LLMs should be used as writing aids, not much more. He argues that marrying current LLMs with tools such as search engines is highly non-trivial. LeCun predicts that there will be a better factual, non-toxic, and controllable system. They won't be auto-regressive LLMs. AR-LLMs make stuff up and should not be used to get factual advice. LeCun warns that only a superficial portion of human knowledge can be captured by LLMs.