I think the idea is that nobody will be using CC in 5 years. If anthropic loses, nobody will use it. If anthropic wins, still nobody will use it! The value is in the model solving problems, and CC is just the hacky vessel for that, not the end goal. If they believe in themselves, polishing the product is a waste of time.
I thought Gemini just started providing citations in the last few months. Are you saying they should have beaten Google to the punch on this? As part of the $500,000 budget?
Text comes in, text goes out, but there's a lot of complexity in the middle. It's not a "world model", but there's definitely modeling of the world going on inside.
I remember a lot of pop sci being centered around "elegance", looking for simple models that are broadly predictive. Newton, Galileo, Einstein, Darwin. Feels like people are leaning the other way now, and seeing reality as messy, uncertain, and multifaceted.
The whole trick is having your cake and eating it too. The LessWrong style exploits the gap between the strength of the claims ("this is a big deal that explains something fundamental about the world") and the evidence/foundation (abstract armchair reasoning, unfalsifiable)
That could explain the "churn" when it gets stuck. Do you think it needs to maintain an internal state over time to keep track of longer threads, or are written notes enough to bridge the gap?
The question is whether the system can be responsible for the process. Big picture, AI doing 90% of the task isn't much better than it doing 50%, because a person still needs to take responsibility for it actually getting done.
If Claude only works when the task is perfectly planned and there are no exceptions, that's still operating at the "junior" level, where it's not reliable or composable.
Yeah, it's bad enough for capable users, but it's a nightmare for old people and the unaware. The online space is full of scams, and there's no real safe haven.
Great article! It does a good job of outlining the mechanics and implications of LLM prediction. It gets lost in the sauce in the alignment section though, where it suggests the Anthropic paper is about LLMs "pretending" to be future AIs. It's clear from the quoted text that the paper is about aligning the (then-)current, relatively capable model through training, as preparation for more capable models in the future.
I think it's more about social stratification than bodies of knowledge. The knowledge is treated as a class signifier, especially by the protanogist. In the bit with the friend, the new training he didn't have was practically useful, but, more than that, it sharpened the gap between the "haves" (went to a good school) and "have-nots".
I don't think it's accurate to describe LLMs as pattern matching. Prediction is the mechanism they use to ingest and output information, and they end up with a (relatively) deep model of the world under the hood.
It's a classic problem when someone reaches the edge of their competence, but still cares about success. They don't see anything they can do differently, and the focus becomes how other people are (apparently) screwing up.
I think self-cataloging is fundamentally masturbatory. On its face, there's nothing wrong with keeping notes or searchable records. But letting the record become the goal - organizing, re-organizing, polishing, theorizing - feels wrong in a way I can't articulate.
Isolated populations "drift" genetically, which is how species form in the first place. Discriminating against people based on those differences sucks, sure, but it's still the ground truth.
I think the idea is that a wargame is more like D&D, and less like the Olympics. It requires cooperation to keep the illusion intact, so there's a limit to how "sweaty" you can be trying to win. After a certain point, you're exploiting the limitations of the scenario rather than the weaknesses of your opponent.
It's technology propaganda. The protagonist is initially skeptical, but learns to accept the wonders of tech as her life is magically transformed by it.
Clearly not true. In 2020, the SEC fined Andeavor for buying back its own stock while negotiating a potential merger. The duty is to affected shareholders and the integrity of the market, not the company.
Don't know how you got this from Matt Levine. Isn't his catchphrase "Everything is securities fraud"?
AI has a lot lower bar to clear to upend the tech industry compared to the global economy. Not being in on AI is an existential risk for these companies.
This guy is a great historian. He keeps gunning for the fundamental question - "What was it like to live then?", and backs it up with rigorous foundations.