nathanasmith

Exactly. To butcher a cliche, if the model has 6 hours to chop down a tree I want it to sharpen the axe for 5 hours.

They begged to be regulated and now they're being regulated. The company doesn't get to pick and choose the exact form of the regulations they get and in this case they got more than they bargained for. Maybe next time be more careful with the messaging.

There won't be any new generation of models more powerful than Fable since the argument against Fable would apply even more. Opus 4.8 and GPT-5.5 is the best we'll ever see from this point forward. Soon low cost Chinese models will catch up to those thereby destroying Anthropic and OpenAI's pricing power which will mark the beginning of the end for them too.

This heralds the end of frontier model development in the US since the same national security argument can and will be made against any model stronger than Fable/Mythos. Squashing the ability of Anthropic and OpenAI to deploy newer stronger models will destroy their valuation so no trillion dollar IPOs either. Low cost Chinese models will soon catch up to Opus and GPT-5.5 eroding Anthropic and OpenAI's ability to charge more. The knock on effects of this are just beginning.

I start out writing most of my terminal applications and utilities in Python but when something hits a performance ceiling I convert it to Go. That's been a pretty good bar for when it's time to use Go for me and so far so good.

> Would I buy clothes [...] from Costco?

Yes. Happily.

The verbs are one of my favorite things about Claude Code.

The thing that bothers me about "warmer, more conversational" is that it isn't just a cosmetic choice. The same feedback loop that rewards "I hear you, that must be frustrating" also shapes when the model is willing to say "I don’t know" or "you’re wrong". If your reward signal is mostly "did the user feel good and keep talking?", you’re implicitly telling the model that avoiding friction is more valuable than being bluntly correct.

I'd much rather see these pulled apart into two explicit dials: one for social temperature (how much empathy / small talk you want) and one for epistemic temperature (how aggressively it flags uncertainty, cites sources, and pushes back on you). Right now we get a single, engagement-optimized blend, which is great if you want a friendly companion, and pretty bad if you’re trying to use this as a power tool for thinking.

I had an old Galaxy Tab S7 collecting dust on the shelf. Since iOS 26 came out I find myself reaching for the Android tablet more and more. First time that ever happened. (Sent from my Galaxy Tab)

Readers would have been better served with the prompts you wrote than the AI generated output.

The person you replied to is in Pakistan.

>You're right - I don't really care if the track playing in my favourite cafe is AI-generated or not. You're not supposed to be emotionally invested into background music

I guess different strokes but some of the best music I've ever been turned on to just happened to be playing in some random cafe or coffee shop. Conversely if the music is bland and uninspired I'm much less likely to go back.

Is there an API for Grok yet? If not that could be the issue.

Unfortunately that wouldn't help as much as you think since talented AI labs can just watch the public leaderboard and note what models move up and down to deduce and target whatever the hidden benchmark is testing.

I had been sleeping on Claude's ability to write books until a couple of days ago I had it write a novel set in the Accelerando universe. It whipped up a very convincing complete multi-Act 13 chapter side plot about humans learning to interact with Economics 2.0. It was quite good though I'm sure cstross would be horrified.

I have a T420 I've been using for years. Upgraded to 16GB of RAM, SSD, swapped the dual core i5 for a 4 core/8 thread i7 (yes, the CPU is in a socket!), and swapped the 1600x900 crappy display for a newer 1080p panel that looks much better. I absolutely love this laptop and am not looking forward to the day when it's too old for the modern web.

For the lmarena leaderboard to be really useful you need click the "Style Control" button so that it normalizes for LLMs that generate longer answers, etc. that, while humans may find them more stylistically pleasing, and upvote them, the answers often end up being worse. When you do that, o1 comes out on top followed by o1-preview, then Sonnet 3.5, and in fourth place Gemini Preview 1206.

If I want to talk to AI I know where chatgpt.com is. I don't need it shoved in my face when I'm trying interact with people on social media.

I like how it cites relevant Youtube videos based on the search and shows thumbnails of the videos in its results. As far as I can tell ChatGPT doesn't do this.

When Half-Life 2 came out it caused me to break a video game addiction I'd had since being a teenager. I was so awestruck by the quality and enjoyment I derived from the game that after playing through it any other game I tried later paled in comparison to the memory. It got to the point that I couldn't make it more than 30 minutes into a new game without losing interest and eventually I just stopped buying games altogether and that was it. I'm still not sure if this is a good or bad thing.

They weren't legally allowed until very recently.

Leave it alone. It's fine the way it is.

I don't know but I have a Galaxy Fold and I hate the tiny bezels it has in tablet mode. Trying not to touch the screen while holding it adds unnecessary cognitive load and just makes it feel fiddly. I also have a previous gen iPad mini and I love the thicker bezels.

My experience using LLMs to learn is similar. When I read MDN or some O'Reilly tome I get a lot of information but it's in the general sense. I can use what I've learned to build some specific project and it'll work but because the book isn't tailored to the specific thing I'm doing, there will often be a much better way. The LLM on the other hand gives an answer as specific as I'm willing to give it context for and since I know software engineering as a discipline I know when the specific suggestion from the LLM is far superior to the more general method learned in the text book.

My main issue with this is the two hosts seem to be a little too in "sync" with each other. Like they're completing each other's thoughts and sentences without missing a beat. It breaks the illusion of it actually being two different people. Other than that I'm excited about the future of this kind of thing.

They would be wondering why it took so long.

I heat water on the stove top which is plugged into a 240 volt outlet.

The system has 512 GB of RAM so while it'll be slower at inference, he really has about 704 GB at his disposal to run the model assuming he distributes the weights across the VRAM and system RAM.

I stumbled on an obscure question years ago on one of the sub exchange sites that dovetailed precisely with something I'd been working on for years. All the proposed answers were subpar so it was like I was born for this particular question. After cracking my knuckles and preparing to reply I realized I couldn't. I didn't have enough karma or whatever. And that was the end of that.

People who order at the counter subsidize those who order with the app. That's basically how it works now.