prodigycorp

I’m typing from my phone and im not going to review the semantics of Anthropic’s storied history of performance issues.

It’s not just ant. There are so many small knobs that providers can claim isn’t nerfing but “load management” or “improving user experience”. One example from OpenAI is reducing juice to reduce time to first token.

Man this was last year and some Claude subreddit drama that I can’t furnish offf the top of my head but maybe one of the historians remember it.

Incorrect.

Anthropic has admitted to nerfing in the past. There have also been inference bugs. On top of that, model performance changes as they move compute to schwaggier providers as well.

Your chart is wrong.

This could and probably is me projecting, but the enthusiasm was tepid. The Opus 5.5 mog cast a pall on the room.

The stuff that came off as potentially exciting was the ultrafast stuff available only to Pro 500 users.

The reset button stuff was an embarrassment. Everything that was released today, in respect to the hype that the team projected on twitter, was a failure. The internal OpenAI slack is a hell of an echo chamber.

API price cuts were obvious once they made their announcement changing how usage is counted.

These moves all make sense when you take into account the enterprise market.

https://news.ycombinator.com/item?id=49889873

This is a bad faith article. The feel of a "buggy mess" is different from what the author described. Every OS is going to have bugs. On top of that, I cannot reproduce a lot of these issues. It's clear the OP is using multiple third party utilities to manage things like the menubar. Try turn them off and complaining.

I feel like the author wanted to write a post like one of those about Tahoe (menu bar, corner dragging), but the stuff they wrote about was all very "who cares"?

Golden Gate has reversed many of Tahoe's regressions and further updates will continue to do so.

  > This one just irks me. Show me one oval camera lens anywhere in the world. I'm waiting.

Well, OP, today is the day you learn about anamorphic lenses

This is disrespectful to the people of hiroshima who had to walk ankle deep through blood and bone just to find their family

Great course, thanks for sharing, going through your slides and papers right now.

I'm guessing the answer to this is that, despite being much more than what codex provides, claude current thinking summaries are not useful for training a frontier model.

btw your comment was immediately marked as flagged/dead. Odd for a high quality comment. I vouched for it.

My main point of disagreement would be that I fundamentally see a different use for a jev sort of model (generalism is applealing), but if you're finetuning, a bert base is not bad.

Some things are just obvious.

People are up in arms about this change but I think the knee-jerk reaction here is missing the bigger picture. API pricing is probably going down as a move to be more competitive in the enterprise space.

The ppl paying the actual bills have all been exploring crappier models to reduce costs and I think that has been strongly communicated to OpenAI.

Edit: People who think this is death for OpenAI need to look at Anthropic’s revenue. It has flatlined. Stingy decision makers are a bigger risk to these companies than stingy consumers.

laya sucks. it's a dead end and for some treat it as a peer of jev or even the qwen jevs. base model doesnt have enough world knowledge to generalize.

It was not about oil. That is a harmful lie. The reality is more sobering and dangerous: it's the neocon agenda that led the invasion. There are people who fundamentally believe in nation building as national duty. That ideology should be addressed and resisted directly.

Distracting people with the oil excuse gives cover to ppl who need it least.

The comparison i'm making is against a training of encoder/small decoder model so I think we agree on most things. I dont think jev replaces the benefit of embeddings nor think they are mutually exclusive. All part of a handy utility belt.

Yes. The difference is you can immediately benchmark and iterate via prompting without having to retrain. It's such a time saver.

People have been making their own calibrated classifiers and decision makers for a very long time. Mine outperform jev. But that's not the point of jev.

how many ms is it for a 32k token prefill?

They make it clear that their edge is in their synthetic data. Train your own jevs miss the point. And if you have that much data, you didnt need jev or these replacements in the first place.

HN's desire to pretend the data pipeline doesn't exist or isnt meaningful is silly.

yeah i know they are summarized/condensed, but they are still an order of magnitude more useful that what they competitors present to users.

latest. click on the down caret next to the session title. hover over transcript view and click on 'thinking'

they dont show thoughts in chat, but they do in code mode.

It's not there in the chatui but claude code shows more reasoning than they need to.

im not suggesting that they should, but it's so much more than the garbage both antigravity and codex show to users. seems like a thing a company paranoid about distillation would reign in.

Can someone answer this: if anthropic cares so much, why don't they meganerf their thinking summaries (edit: in Claude Code) the way OpenAI does? OpenAI's thinking traces so much that the Codex app doesnt even show them.

Article must be wrong.

  > During operational flights, SpaceX plans to bring both Super Heavy and Ship back to the launch pad, where they'll be caught by the tower's "chopstick" arms. The company has already done this three times with Super Heavy but has not yet attempted it with Ship.

Distillation: Stream starts in about 1hr. Payload is 26 starlink v3 satellites.

Launch window is 8:15 AM to 9:30 AM eastern time.

Superheavy will land 7min after launch. Starship lands 9hrs post launch.

Edit: Both will be splashdowns (originally said they'll be caught)

https://x.com/i/broadcasts/1qxvvenMPMQxB

https://www.spacex.com/launches/starship-flight-14

It's very good, yes, but I expected it to produce midjourney type results out of the box. That did not happen. The models are definitely granular stuff now though. They must be training off a ton of digital artist stroke data now.

But it other cases, like the music videos, much of the magic is done by access to elevenlabs and suno apis.

Edit: just saw your edit about the pdoom video. Can you share how you prompted it? Would be helpful to know.

What we know after the openai incidents is that the RSI process involves models having access to user rollouts via tool calls. What's considered crappy training one year is another year's kompromat!

Opus 5.5 is a good model, but I've tried to understand the extreme hype about it on social media about Opus' ability to do 2d work, as we got with Astra doing 3d work. In both releases, the models required extensive access to third party apis to generate assets for it, and a lot of the models work was essentially coordinating everything.

There's so many "x generated this in one shot, this is agi" stuff that gives you the impression that you can vibe operate modern models the same way you operated last year's models. There's so much more to it than that. It requires you to put a faith in the leap in the capability of models, one that would've surely been a waste of time in previous models.

Not sure where i'm going with this other than I think most can relate that it's exhausting keeping up with. I cant imagine what it'd be like parenting a kid that went from toddler to puberty in the span of a year and planning for them to go to college the next year. This industry is moving so fast that it's becoming fact that it's the user that's "holding it wrong" every six months.

The tldr is that muse is heavily inspired by openclaw and should be considered FB’s version of it.