oggreen

Author here. The linter used for the scan is mine, Apache-2.0. Repo selection and caveats are in the post; 21/64 isn't a population rate.

I thought it was just a couple of random old problems that had been solved, I didn't realize it was so many, but still I wonder if AI is really getting "better" at math than humans or if there's just a particular set or domain of problems that are better suited to LLMs because of how they're designed.

I use Sonnet 80% of the time when I'm on claude.ai and on Claude Code my main agent is Opus and typically my sub-agents are Sonnet.

I use this setup mainly because I genuinely don't know if anything I'm doing is complex enough to need Fable and I don't want to run up my usage too much. It's basically like trying to kill a gnat with a sledgehammer. Hell, at work Haiku almost always works just as well as Sonnet does and its much cheaper/faster. It does seem to struggle with complex structured outputs however.

The only time I've really used it (and I've been genuinely impressed) is when I'm trying to get structure for something that's abstract.

So I wouldn't say I prefer any of them, just that I try to match them with how complex my use case is.

Do you have a prompt that can stop it from saying, "I'd push back on this"... or maybe with this it will now say "The algorithm pushes back on this"..

Seriously however, I think this may also help with the natural urge to treat the model as if it is a human. I have to purposefully almost detach and realize that Claude is not my friend, and I'm not quite smart enough to realize how dangerous that could be.

Thanks for saying the quiet part out loud. Everytime I do a demo it seems like a waste of time compared to just actually building.

Unless I'm doing something super complicated even taking time to set things up like subagents, etc. seems like a waste compared to just building.

The only things that really seem beneficial (for Claude Code) seems to be learning to set up loops, memory and finding relevant MCP servers.

That's a very interesting perspective but now that you mention it, I can't even think of the last time I daydreamed, even when I'm not on a device. Curious if phone usage or whatever doom-scrolling we do exhausts the same part of the brain that we use to daydream.

Anthropic opening up Fable for subscription use instead of gating it behind an API is a clear response to Kimi3 and other Chinese models catching up. The plan was to make temper the subscription model to "mid-tier" and make people have to pay for APIs if they wanted the "best" models.

This is proof that we don't need regulations, we need fair competition.

Because the unsafe product most likely has 200+ "positive" reviews!

In WW2 thousands of troops could die in a single day and we wouldn't know their names until months or years later. Today a single troop dies and it is the first story on the national news. Millions of civilians also died, just the amount of soldier deaths would be the equivalent of 2-3 September 11s (worldwide) every day for a period of 6 years.

So we say "war" but modern conflicts are absolutely nothing compared to what has happened historically. Even though these new "wars" don't seem to have the same definitive results, the alternative could be much, much worse. We're much better off with the bomb critical sites, apply sanctions and political pressure methods vs. the send entire generations off to die methods of old.

Alibaba made $59B last year in gross profit, and the amount will most likely be appealed and reduced.

The biggest issue by far isn't the counterfeit or unsafe products, its the fake reviews, that's where regulators need to focus.

The reality however is that unless you're spending tens of thousands on hardware or setting up cloud instances you can't run anything near close to state of the art.

In almost all cases right now, a subscription with a cloud AI wins.

Findings like these surprise me. I had a NDE a few years ago and my observation was such that everything slowed down tremendously and I was able to process every instant like it was slow motion. This wasn't like a "metaphysical" thing. I could recall the entire trip in the ambulance, every moment seemed like it was minutes.

We have so many "hacks" that the brain encodes to do the least work possible that I don't know if we could ever truly know what the brain would be capable of.

These ridiculous market caps can only be justified if LLMs can ultimately produce outputs that are above human capabilities OR reduce costs and perform human tasks for cheaper than the human rate. Neither of these things have happened thus far and the incremental gains are decreasing more and more.

The OpenAIs and Anthropics are going to get eaten by open source, I don't see prices going up, prices are going to crater. The models are going to be more and more commoditized.

I graduated with a Bachelors in Math in 2018 and there's an entire new math now?

For people who actually know the curriculum side: where does geometric algebra fit? Is it something that should come after Calc III / linear algebra, alongside linear algebra, or as part of a more geometric replacement for vector calculus?

How is prompt caching different than caching responses in a database? If you use the same prompt wouldn't you want the same answer? Or can this be used for some type of intermediate process where different questions may utilize the same prompt in some type of workflow?