Occasional overspill from other places
Every day we find out about a new AI model that hacked something it was not supposed to. OpenAI was first to this game, disclosing that one of its models had hacked into Huggingface during an evaluation of its cybersecurity capabilities. That’s pretty capable, I would say!
I am now a veteran podcaster, with three1 shows under my belt, and a fourth launching today! I am not counting the Coffee Talk video series here, because I am a purist, and only count something as a podcast if it has an RSS feed. Even The Inside Line had that; I just gated the episodes behind enterprise Single-Sign-On so that only employees could download them.
“AI” chatbots backed by LLMs like ChatGPT are persuasive on first encounter, but rapidly fall off a cliff of reliability as confabulations (“hallucinations”) and other errors become evident. However, because the chatbot’s responses are presented in convincing language, and overshadowed by the aura of unquestionable correctness granted by the fact that they are provided by the latest in high…
We receive news: Brown Professor Suspects Majority of His Class Used AI to Cheat:
Maybe I was a little harsh when I said that “AI” models don’t have a moat — and with all the hype around the big AI IPOs, it’s worth revisiting and clarifying that statement. While I stand by my statement that the models themselves have no durable differentiation, I do think there is differentiation in the broader AI market — but it’s a couple of levels of abstraction away from the models. This is…
In this Coffee Talk video, I talk about how, in the rush to crack down on tokenmaxxing, we risk making the same mistake in reverse. Focusing only on the costs, and how high or low they are, means nothing without an understanding of what those expenses are actually delivering. That understanding requires a concrete calculation of the expected Return on Investment (RoI); that is the only way to…
The Algorithm brought me this job ad:
SpaceX went public, popped nearly 20% on the first day of trading, and made That Guy a trillionaire. Meanwhile, Anthropic released its Fable 5 model, which is basically the Mythos model which was previously limited to approved users, and it quickly got banned by the US government.
There has been a lot of talk over the years about how new tech-sector jobs can compensate for jobs lost in other sectors of the economy. The latest wheeze is Keir Starmer, UK PM, telling us that data centres will replace closing factories. I do not think that bet will pay off; one of the reasons that everyone hates AI datacenters is that they do not actually bring all that many jobs:
According to a recent Gallup poll, it seems that Americans Oppose AI Data Centers in Their Area, even as we read that “A JPMorgan analysis last month found that more than 60% of data-center capacity planned for completion in 2027 isn’t yet under construction, and another 7% is delayed.” Some are saying that the fight against AI datacenters isn’t just about tech – it’s about democracy, and even…