Speculation: open models is what will kill Anthropic and OpenAI. Hyperscalers can run the models without a licensing fee. Apple can make them smaller and put them on the device.
The frontier models are an edge and a liability. They're astronomically expensive to train. Without them, their models will fade into obscurity. Their marketing depends on people believing the models are meaningfully different, as people have sweatily argued on this forum. Personally, I'm not convinced there's much of a difference between these models at this point. The harness is what takes these random and hallucinogenic models and make them into something deterministic and useful.
Lots of good advice here. I also read one book a week and wanted to add my POV. I don't put any special effort into reading books during my downtime, as the OP describes here. I just read when I feel like it and before bed.
Here's my take as someone who was using Claude everyday for every little thing and now I've deleted my account because I didn't like what it was doing to my mind.
What goes totally unmentioned in the article is that this feature is designed to help mitigate dark usage patterns. My major concern with chatbots is excessive usage can lead to AI psychosis or negative rumination (depression). A feature that makes user's more self-aware of their usage patterns is a good thing. Making the user more self-aware is a necessary first step that will precede any kind of intervention to reduce their reliance.
Where this fails is it frames the intervention as a moderation problem. It may seem counterintuitive, but moderation takes more self-control than elimination. If you struggle with your relationship to LLMs, then every time you make a choice whether or not to engage with a LLM is an opportunity to struggle. The more you struggle, the worse you feel.
Obviously Anthropic cannot advocate for churning from their product. The psychological stickiness of their product is its primary selling point for investors. When they say "set quiet hours and breaks" it frames this as a user problem. Just get good bruh, it's not that the technology hallucinates or is sycophantic, or basically designed to be a AI girlfriend / boyfriend, it's a skill issue. Rather than a technology being applied incorrectly and a company floundering to hook users before they try to jack up the prices to stay solvent.
I find the "AI Fluency" course particularly ridiculous.
> Build AI skills that support your original thinking
This is a straight-up lie. When you outsource your thinking to Claude then your ability to produce original thought degrades. The whole framework for using LLMs is the sort of thing you see from tech bros on Twitter trying to sell online courses. It reminds me of the intellectual yet idiot essay by Nassim Taleb[1]. Don't let Anthropic tell you how to think under the guise of doom trolling[2] and tech bro "thinking frameworks". Think for yourself!
Thanks for this. I've been trying to learn this writing thing on top of my full-time job and it's been a struggle. It's helpful to hear how you managed it.
The "hiding from researchers" framing is particularly bad. The parsimonious explanation for why a model produces different outputs when it detects eval contexts: eval contexts appear differently in the training distribution and the model learned different output patterns for them. No theory of mind required. Occam's razor.
The agentic behaviors emerge from optimization pressure plus tool access plus a long context window. Interesting engineering. Not intent.
People are falling for yet another Anthropic PR stunt.
I see a lot of people are confused about the electricity claim so I'll elaborate on it more. The assumption I'm making here is that on device people will run smaller models, that can fit on their machines without needing to buy new computers. If everyone ran inference on their machine there would be no need for these massive datacenters which use huge quantities of electricity. It would utilize the machines they already have and the electricity they're already using.
People are making a comparison of the cost per inference or token or whatever and saying datacenters are more efficient which makes obvious sense. What i'm saying is if we eliminate the need for building out dozens of gigawatt datacenters completely then we would use less electricity. I feel like this makes intuitive sense. People are getting lost in the details about cost per inference, and performance on different models.
LLMs on device is the future. It's more secure and solves the problem of too much demand for inference compared to data center supply, it also would use less electricity. It's just a matter of getting the performance good enough. Most users don't need frontier model performance.