This post was originally published on here
American Computer Scientist David Magerman has a message for almost everyone on artificial intelligence (AI) or rather the AI doomsday prediction. David has been at the forefront of computer science since the 80s and throughout his career has been deeply involved in early efforts to apply machine learning and data analysis technologies to capital markets investing. Magerman is the man who helped build the trading systems of Renaissance Technologies, widely recognized as the best quantitative hedge fund management company in the world. Magerman holds a PhD in Computer Science from Stanford University where his thesis on Natural Language Parsing as Statistical Pattern Recognition was an early and successful attempt to use large-scale data to produce fully-automated syntactic analysis of text.Presently, David Magerman is managing partner and co-founder at Differential Ventures, an early-stage fund that invests in data-focused entrepreneurs building enterprise technology. Magerman is among the scientists who can be said to buck the consensus on AI. In a podcast, earlier his year Magerman said that AI is a machine gun we are giving to kids and definitely a bubble. He is also no fan of Sam Altman’s OpenAI, in the same podcast, Magerman said, “OpenAI isn’t a a functioning business. The business model is so demonstrably a Ponzi scheme…” In a recent long note on LinkedIn, titled ‘Something Big is Not Happening’, Magerman writes why all the forecasts about AI taking over all humanity is wrong. He says, “Please, please, please don’t take this essay too seriously. Don’t believe hype. Look at facts.”
Here’s what can be called a ‘comforting post’ from David Magerman
There is a dramatic, viral, largely AI-generated essay going around, claiming we are already in the midst of some apocalyptic age where AI is going to transform humanity in some irreparable and harmful way. Please, please, please don’t take this essay too seriously. Don’t believe hype. Look at facts.LLM-based generative AI solutions are still barely being used in production environments at scale. Most use of LLM-based AI in major corporations is in limited pilots and non-mission-critical internal tasks. And most of the pilots are failing.There are fantastical reports of LLM-based AI solutions doing unbelievably sophisticated things. Many of them are easily debunked (see moltbook). Some are not so easily debunked, but they are also unreproduceable and unproven.Code-generation is one of the most effective use of LLM-based AI, mostly because LLMs are incredibly good at memorizing their training data, and the open-source community has contributed coding solutions to a broad class of problems, and those solutions were used as training data for LLMs (arguably against the license agreements of the code, at least in spirit). But it is unclear how well LLMs can solve new and complex problems. Unproven anecdotes from conflicted parties are not clear evidence.A lot of rich people and valuable, powerful companies have a lot at stake on everyone continuing to believe that LLMs are at the center of a new world order that will transform EVERYTHING. As a manager of a data-focused venture capital fund, I would benefit a lot from this fictional future as well. But it’s just that: fiction.Transformer models are a generational scientific improvement in pattern recognition and machine learning which will have long-lasting implications in how humans use computers to translate data into better decision-making. But that’s it. Don’t get me wrong: that’s a lot. But that’s it. Everything else is just marketing hype meant to preserve bubble valuations until the major stakeholders in AI-focused companies can figure out how to maintain some of their value or get out before their investors do.Consider the source of everything you read about AI (including this note) before believing it. Try to find as much objective facts as you can about how AI is really being used in the real world, and judge it based on scaled solutions, not anecdotal usage and unverified screenshots of outlandish representations of the supposed power and intelligence of these systems. Presume everyone on the internet is trying to sell you something, and make sure you have enough objective evidence before you buy it.Caveat emptor.
