conradkay

6 ms·

31.
by conradkay 4mo ago
My guess is that they liked the status quo with Project Glasswing and didn't want Fable to be public, especially if anyone is jailbreaking it into Mythos and using it for cyber But then it backfired spectacularly and now it seems they
32.
by conradkay 4mo ago
I'd guess it's absurdly good business wise, given API pricing is like 10-30x as much
33.
by conradkay 4mo ago
It's certainly worse news for Anthropic than other labs since it's not completely random, and there's people in the administration (e.g. David Sacks) who don't like Anthropic -- perhaps seeing them as an enemy
34.
by conradkay 4mo ago
Seems like estimates are that 70-85% of their revenue comes from API usage/pricing, so some users switching from Opus to Fable for that would've had a big impact Then there's people switching from GPT 5.5 or upgrading their s
35.
by conradkay 4mo ago
Does low/medium effort fix it for you? Seems like Fable 5 low can outperform Opus 4.8 high/xhigh often, and uses a lot fewer tokens
36.
by conradkay 4mo ago
Looking at page 255 of the model card ( https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c3... ) it might be much better on all dimensions (speed, cost, quality) to just use Fable 5 on low/medium effort than swi
37.
by conradkay 4mo ago
Looks like it'll be more like $2t https://polymarket.com/event/anthropic-ipo-closing-market-ca...
38.
by conradkay 4mo ago
I suspect it's less true now that synthetic data has worked so well, and multimodal doesn't seem to transfer as well as many would've hoped Claude Code and Codex are a big advantage, vs Gemini CLI which might be killedbygoogl
39.
by conradkay 4mo ago
Yeah it's very murky, but if they're arguably profitable under some definition of profitable, then it's ridiculous to claim there's "no path to profitability" Especially if you assume that 6 months ago they wer
40.
by conradkay 4mo ago
I think there's just less time between model releases now
41.
by conradkay 4mo ago
Probably not till mid June
42.
by conradkay 4mo ago
That feels really hard to argue now that Anthropic/OpenAI are so much bigger How the hell do you crush a ~1T company on the one thing they have all their focus on?
43.
by conradkay 4mo ago
https://techcrunch.com/2026/05/20/anthropic-says-its-about-t...
44.
by conradkay 5mo ago
I do find it somewhat concerning that the incentives aren't aligned, but as long as things live on your computer it should take just a prompt to migrate to something new And Codex is open source
45.
by conradkay 5mo ago
GPT 5.5 is maybe 4x the size of v4 pro, hard to compare price since cache hit is basically free with Deepseek, but 40x cheaper (with their 75% off) seems about right. So ballpark same price per parameter as Simon.
46.
by conradkay 5mo ago
You'll have access to the same models as your hypothetical attackers, and a big advantage if only you have access to the source code
47.
by conradkay 5mo ago
https://www.techinasia.com/news/chinas-deepseek-eyes-10b-fun... Would be weird if they're raising $10 billion after spending only 0.3 https://newsletter.semianalysis.com/p/deepseek-debates Pr
48.
by conradkay 5mo ago
That's one way to look at it, though it feels like you could say the same about the dot-com crash or 2008 which isn't too helpful. At the very least (extremely high-paying) jobs can be actually lost
49.
by conradkay 5mo ago
The numbers for Meta are pretty misleading, I assume the $3 billion is something like direct generative AI revenue? Sure they're torching money on building consumer LLMs, but they seem to be doing very well optimizing things like ad ra
50.
by conradkay 5mo ago
It's still very contrarian to expect GPUs won't depreciate rapidly. Yes 3090s were a good investment then, but way worse than just buying Nvidia stock directly
51.
by conradkay 5mo ago
It'd be way slower since you'd be doing that work every token
52.
by conradkay 5mo ago
Mine's set to 1 day (seems to be enough from all the cases we've learned about), I got you. Also seems like this attack and most others were caught by automated tooling from 3rd parties
53.
by conradkay 6mo ago
Yes, https://x.com/ajassy/status/2048806022253609115 (Andy Jassy) "Very interesting announcement from OpenAI this morning. We’re excited to make OpenAI's models available directly to customers on Bedrock
54.
by conradkay 6mo ago
They're months behind now and have very low market share, so as long as they stay months behind the duopoly/triopoly can hold.
55.
by conradkay 6mo ago
They mean Anthropic secondaries are trading at more than a trillion (allegedly)
56.
by conradkay 6mo ago
Recursive self-improvement is one argument. Otherwise winner takes all seems much less likely than a OpenAI/Anthropic duopoly. For the best models, obviously other providers will have plenty of uses, but even looking at the revenue rig
57.
by conradkay 6mo ago
> I feel for 40B you can get a lot of very smart people and a lot of very good hardware to outcompete it Nah, see Meta
58.
by conradkay 6mo ago
Was 4.7 distilled off Mythos (which got 77.8%)? Interesting how mythos got 82% on terminal-bench 2.0 compared to 82.7% for GPT-5.5. Also notice how they state just for SWE-Bench Pro: "*Anthropic reported signs of memorization on a sub
59.
by conradkay 6mo ago
They say latency is the same as 5.4 and 5.5 is served on GB200 NVL72, so I assume 5.4 was served on hopper.
60.
by conradkay 6mo ago
Codex has a lot better limits, and 5.5 will be out soon

More ›