Artificial Intelligence– Spray and Pray

Irrational exuberance is still with us.

• The valuations being attributed to certain AI start-ups who have no products and talk like they don’t care about profit continue to defy reality as they are very unlikely to uncover the secret of intelligence upon which their valuations depend.
• The two companies in question are Thinking Machines Lab founded by Mira Murati (ex-CTO OpenAI) and Safe Superintelligence founded by Ilya Sutskever (one of the founders of OpenAI).
• Thinking Machines Lab is raising $1bn at a $9bn valuation while Safe Superintelligence is also raising $1bn but at a valuation of $30bn up from the $5bn valuation at which it raised in September of 2024.
• Both of these founders left OpenAI over concerns around the safety of AI in the case of Ilya and, in the case of Mira, I think the pivot towards being a for-profit entity was the main reason.
• Both Ilya and Mira have started their own companies to focus on the safe use of AI superintelligence (IIya) and creating AI and making it available to as many people as possible (Mira).
• Neither of these companies appear to be particularly concerned with making any money and seem to be more focused on making AI to benefit all of mankind.
• This is a noble cause but the present value of a company with no net cash flow to shareholders is zero meaning that these companies should be funded by philanthropic organisations, not venture capitalists.
• If there is no net cash flow to shareholders, then the only way that these companies are worth anything to their owners is if they invent super-intelligent machines and then sell or license the IP to 3^rd parties.
• If either of these companies invent super-intelligent machines, then whoever wins that race will be worth far more than their current valuations assuming that they do not give the technology away to benefit mankind.
• As a result of these companies being founded by OpenAI alumni, the assumption is that they will invent super intelligent machines and not give the innovation away.
• These are two big assumptions, and I have serious doubts about anyone’s ability to create AI super intelligence for as long as the machines have no understanding of causality.
• I also have serious doubts about whether these machines are capable of reasoning in the way that the creators of these machines would have us believe.
• They can do the complicated stuff really well but quickly fall over on the stuff that 5-year-old children can do, implying that they are not really reasoning but merely offering a sophisticated simulation of it.
• This is also crucial because the ability to reason is the first step towards true artificial superintelligence and is one of the key signals that I am looking for to indicate that progress is being made.
• In the absence of this I continue to think that we are no closer to artificial superintelligence than we were 10 years ago and as a result, there is a correction coming.
• These sorts of companies will not survive a correction as they have no product revenue to fall back on when the market realises that artificial superintelligence is not coming anytime soon and the funding stops.
• In the meantime, the valuations being attributed to Safe Superintelligence and Thinking Machines Lab are a sign that the market still thinks that super-intelligent machines are close to becoming a reality.
• This means that anything with a hint of credibility gets funded which is precisely what happened in the internet bubble of 1999 and 2000.
• I continue to think that when the correction comes it will be much milder than the multi-year correction observed between 2000 and 2005 as the AI being created today has real use cases and will generate real revenues.
• Hence companies like OpenAI, Anthropic, Safe Superintelligence, and Thinking Machines Lab are unlikely to survive and will be acquired by companies who have the resources to finance them through the correction.
• This is why, I have no desire to jump on the bandwagon and remain comfortable with the adjacencies of inference at the edge and nuclear power.