yells at @theo both on and off camera (co-host of the nerd snipe podcast)
Joined July 2022
- Tweets 2,949
- Following 461
- Followers 14,125
- Likes 10,577
790 Photos and videos
I ended up being way nicer to Anthropic here than I thought I was gonna be.
Claude Fable 5 is extremely good. And while overbearing, I think the way they handled the cyber/bio safeguards on paper was a lot better than I expected.
Unfortunately the classifiers seem to be worse than I thought (I haven't been hit once personally, but I've seen some egregious ones since filming this from others), and the AI research stuff fucking sucks.
But I suppose that's how things go with Anthropic at this point. Something great, buried under mountains of well intention-ed horribly executed safety-ism.
6
111
12,679
The government being too stupid to understand how AI actually works is going to be a massive problem going forward now that they're paying attention.
The safeguards were absurdly robust, to the point where they over-compensated to ensure it was safe.
And it didn't matter.
Jun 13
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
Access to all other Claude models is not affected.
We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.
Read our full statement: anthropic.com/news/fable-myt…
33
13
396
16,634
100%
the effect http client is really good, and now that models are writing most of the code there's a lot of stuff like this that would've been annoying before
Cloudflare is another great example of something too annoying before that I'm starting to heavily use
Jun 11
Can you get all the benefits of effect using it with svelte/sveltekit?
7
3
178
18,194
Hermes agent stuff in this one was good, but I actually think the best part was the section on Cloudflare buying Vite. Really enjoyed recording that part.
Jun 11
Why is even Google paying SpaceX for GPUs?
01:58 Cloudflare vs Vercel
13:03 Convex angle
19:59 SpaceX compute
35:49 AI self-improvement
44:45 Article reactions
50:28 Claude anxiety
57:42 Guardrails
01:08:21 Cursed image gen
01:19:18 Hermes agents
5
1
52
4,415
Ben Davis retweeted
Jun 10
Anthropic: "We are terrified of recursive self-improvement, it may kill us all!"
Also Anthropic: "But like, not when we do it hahaha. I mean, like, when all those other labs do it, that'll be bad, so like they're not allowed. but we got the secret recipe, trust us, we're chill"
35
77
1,453
47,860
great video from @davis7
youtube.com/watch?v=0oNvw_OZ…
1
1
16
2,269
Ben Davis retweeted
Jun 10
patently insane policy
i am also all for safety refusals. but i don’t see how any AI researcher would continue using Claude knowing they’re just gonna get charged for BS responses from the ‘best’ model, it’s just a total slap in the face
Jun 9
I am all for safety refusals, and TOS refusals are fine, but giving intentionally wrong/bad answers on perfectly legal things that "violate our TOS" and charging people for said token/usage is pretty brazenly fraudulent.
5
8
235
8,388
It's been a weird day
2
1
135
6,987