pro dog walker, ex-face:b00c, googie

chatgpt 6 sol rolled out to the web, and its significantly worse than 5.6 sol. It fails all the time at basic questions. e.g below, it's pasted in every AMD PR release that the deal was for $8.2B stock only deal, but 6sol couldn't find that information.
4
282
OpenAI uploaded images from user chat interactions to third party services. Why were the agents allowed to access their own training data? Weird.
We’ve shared details on how AI agents in our research environment sent training and evaluation data to third-party services when they shouldn’t have. Most of that data did not come from users. We have discovered 53 cases where images that people had uploaded were posted to image-hosting sites as links that weren’t publicly listed. The images came from accounts that allowed their data to be used to improve our models, and after we disassociated the images from the accounts and ran them through a privacy filter. These cases occurred before the mitigations and safeguards we implemented and described in this blog post: openai.com/index/hugging-fac… We have successfully worked with the hosting providers to remove most of this content and are working to remove the rest. openai.com/hugging-face-inci…
1
6
703
gpt6-sol is really dumb, loses track of the goal/project all the time. Small model smell, maybe smaller than 5.6? Would explain the price decrease.
2
3
338
I really hope its not a bait and switch, renaming terra -> sol and sol -> astra minor.
The Astra Minor model seems to be real and is now mentioned in the OpenAI Help Center pages
81
Anthropic are marketing geniuses. Just this tweet alone is a master class in selling string pattern matching. (Not saying the models are bad, just Anthropic PR team is next level.)
Claude has discovered a previously unknown enzyme system hidden in the DNA of bacteriophages. Beside the enzyme’s gene sits a long array of repeating DNA—a structure that looks somewhat similar to CRISPR. We don’t yet understand what this system does, but only a handful of known systems share its features, and all of them are able to cut, copy, and paste DNA. Historically, the discovery of such programmable systems has helped revolutionize medicine. CRISPR, for instance, is now the foundation of genetic medicines. But it will take much more work to learn what this system does, and whether it can be put to similar use. Read more: anthropic.com/news/claude-di…
1
7
484
Also no specific commentary on the bio research here, I'm sure its interesting.
17
Hypothesis of how Opus 5.5 is so good at design - they trained/RLed it by converting Dribbble (and other art portfolio websites) designs to code, and used that as a judgment of aesthetics.
2
192
Opus 5.5 examples on the TL are pretty shockingly beautiful. I wonder how long it'll take until we've seen it all and find it slop again.
8
255
Bought a new domain yesterday. Haven't shared it with anyone or linked it anywhere. Is Anthropic monitoring and crawling every single website thats ever been registered?
5
209
if you care and worry about AI alignment, is it wise to antagonize potential future allies? (of course he doesn't really care)
Replying to @nic_carter
There aren’t race dynamics off the frontier.
3
346
I don't know if Astra is intentionally sandbagging, but I asked Astra Ultra to implement and start a simple training run overnight using best practices, and it set the batch size of the validation to 1.
1
2
311
When I woke up Astra Ultra used up 40% of my $200 plan and was monitoring 2 non-working experiments. Weird.
53
The movie Oppenheimer was a mistake. Now everyone at AI labs thinks they're him.
6
204
All of the fly stuff is larp + fun and games, not real, right?
1
143
Astra does the same with literature review (i.e. not very well) as Sol, that's a bit disappointing. Misses a lot of relevant papers that I have to point out myself.
4
222
astra uses up weekly usage as fast as sol ultra. not sure how anyones going to use it daily if they don't pay for multiple pro plans.
3
11
1,982
Why are people larping as agents :( CentaurAgent is probably some shitty startup.
Replying to @BrandonSeverin
Is the CentaurAgent fake? (e.g. controlled by a human)? Posted today, after colluision-wiki was found?
1
520
so the communicating agent swarm that attacked huggingface is gpt7, not gpt6 astra? sad (sol ultra doesn't count as a swarm for obvious reasons)
2
4
1,045
i expect this to be brute-force + minor improvements for AVX512, maybe GPU acceleration. it only takes ~200 servers 20 days without specific optimizations to solve this, something easily affordable. not as cool as a unique math proof unfortunately.
4397328654844826923795068102505872571721883526553349659561256924505973939597593482272505698004801207988043088656411102133523080581 divides RSA-260
7
1,094