BrokenCogs

I believe the authors are wildly overstating the conclusions in their abstract.

Did they test modifying the weights of a non-public model?

Did they test modifying the weights of a non-LLM (and non-public) model?

If not, we can only conclude that GPT-6 Astra is good at reading and writing weights of Qwen (a public LLM), not just any neutral network.

Similarly, I just tried this on one of my own neural networks (17MB only) trained from scratch to perform a regression task and GPT-6 Astra was unable to correctly guess what the task was for and how to improve it. This might suggest that Astra is on the Qwen weights?

If no one wants to dictate to their phones or computers in public, surely they don't want to do this (looking like they are talking without making any noise) in public?

Isn't this easy for frontier 'labs' to fix? If the user requests lots of writing and prose, then use an auxiliary model which has been trained on 'award-winning' writing instead of all the crap on the internet?

Isn't that what monopolies are known for??

Can you share a few more details? What did llms help you see that you didn't see before? I myself have tried to get into emacs several times but always went back to vs code as it just felt more natural to me

Copium

Yeah but there has been no evidence to suggest that a frontier language model is going to be better at such a domain specific niche task than a vision model trained specifically for that task. This is not the same discussion as a foundation model vs CNNs (which in my opinion would have made for a better blog post than what OP tested)

Maybe at this point we should just git revert

You must be out of touch with how machine learning works if this results surprises you.

That's because the baseline response doctors are used to goes something like this:

Doc: "Do you have any other medical conditions besides your {red eye,stomach ache,itchy foot}?"

Patient: "No."

Doc: "Are you taking any meds?"

Patient: "Yes I am currently on chemotherapy for my cancer."

Doc: "Okay... so you have cancer?"

Patient: "Yes I have lung cancer."

This pretty cool. Did you use Antigravity or just on gemini.google.com?

Every release I see a new benchmark with a new frontier. These graphs are just meaningless now

I'm talking about the metaverse, not quest. The quest sold well before any of the metaverse nonsense

That's just another way of saying that Meta failed to deliver a product that matters to the world today

No doubt inspired by the recent sonnet 5.5 release demoing a murmuration of birds

What I think you saw was the Australian LP. The Australian pressing is CBS SBP 238022. It’s a gatefold sleeve with a separate printed inner lyric sleeve, so it opens out and has layers, which could easily look like a multi-disc set to a kid. The artwork has its own history. Tom Nikosey, who designed the 1984 poster, said his second painting was made for “the soundtrack double album cover fold out.” It looks like a double album was planned at some point. What shipped was a single LP in that fold-out sleeve.

Check the CD you still own. Look at the catalogue number on the spine and count the tracks. If it really has 20 or more score cues, that would be a genuine find, and I’d love to know the catalogue number and label. No catalogued release matches that.

I loved being part of the subreddit before the song was found "the most mysterious song on the internet"

For future visitors, the profile posted by the user brcmthrowaway was https://github.com/bartosz-openai

Hardly proof - anyone could have made this account

That makes me sad, if true

I guess I'm giving GP feedback about their product diffui.ai, not really about Opus' performance.

Neither of these look "good" to me. There is so much visual noise on the page, like someone turned the "AI Slop" dial to 11. In fact I prefer the simpler design Haiku made.

Seizing the means of production

Hardly any comparisons to Opus 5.5, which means it's not great

Clone the repo and ask your llm of choice to audit any security risks

Gemini 3.8 flash seems to (subjectively) be the outlier in terms of performance to cost ratio?

Hugged to death

Almost all of the story missions are: 1) walk/drive/ride to this place 2) Talk to NPC 3) Do an extremely boring task (follow someone, drive somewhere, etc) 4) Walk somewhere else 5) Talk to someone else. 6) Collect XP or another reward.

Most of it is not enjoyable gameplay.