Der.dev 🔥🛠️ retweeted
And so it begins: Anthropic just rang the opening bell at Nasdaq. This is going to be a record IPO.
Der.dev 🔥🛠️ retweeted
Shoutout to OpenAI for allowing me to be so absurdly transparent about, well, everything.
Few companies have the balls to give early access to people like me. Even fewer do it with no strings attached.
I've been accused of being an OpenAI fanboy in the past. As silly as it may sound, I think this video shows why.
They listen. They learn. They improve. And most importantly: they've never once asked me to hold back.
They let me share how I really feel, and they take it to heart.
A lot of early access testers were unhappy with GPT-6.2 Astra. None of us were scared to say how we really felt.
The "negative" wave yesterday shows two things:
1. The model kinda sucks to use (sorry, it's true)
2. OpenAI does not restrict their EA testers in any way at all
Seriously. Mad respect to the team for taking this on the chin and letting us be so honest. It's not the norm, it means a lot to me. I hope y'all can take a second to appreciate it too :)
Der.dev 🔥🛠️ retweeted
GPT-6.2 Astra is pretty good at hard things. Here the team shared some results on AMD and Jalapeño chip optimizations.
I do get the feedback on its communication style and we're working on that, but for really the hardest tasks in the world it does seem to be in a league of its own at the moment.
Working with @AMD, our models identified up to 14% PPA improvement on flagship GPU IP in less than five months—building on designs AMD’s engineers had already spent years optimizing.
The same approach helped improve Jalapeño, our first custom inference chip, with gains verified on real silicon.
New evaluations also show our models getting better at certain chip design tasks.
We’re sharing these results and how we’re working with @Synopsys to bring our model capabilities and their trusted EDA design and validation tools to more chip designers.
openai.com/index/advancing-c…
Der.dev 🔥🛠️ retweeted
If you have threads like this or any other code quality issues you spot in gpt code, please do /feedback and describe a comment explaining what you see is wrong with it. Reply to this thread.
GPT's defensive programming isn't safe, it just hides your bugs.
When something goes wrong, good code should break loudly so you notice right away and fix it. GPT does the opposite. It adds "just in case" backups everywhere, so when you make a mistake the code doesn't break, it keeps running with the wrong values, and you only find out much later when things start acting weird.
It's like a smoke alarm that turns itself off so it won't bother you. Feels safer, until the house is on fire. I hate it !
Der.dev 🔥🛠️ retweeted
Gonna take a bold creative risk and admit that GPT-6.2 was an enormous step-up for my own work
Der.dev 🔥🛠️ retweeted
GPT-6.2 Astra is the model that solved Navier-Stokes. Do with that information what you will.
Replying to @davidiach
Yup it is the model that's good at math, and I'm looking forward to all the breakthroughs people find with it!
Der.dev 🔥🛠️ retweeted
Epoch says GPT-6.2 Astra continues the trendline above the old-world reasoning era, supporting my argument that we have been on a new trajectory since September, which will eventually become visible and undeniable.
GPT-6.2 Astra scores 174 on the Epoch Capabilities Index (ECI), on par with Fable 5.2 and joining it above what the pre-Astra trend for reasoning models would have predicted.
This is based on testing we did prior to launch using only our own benchmarks. The ECI is continually updated as more benchmark results become available, and so these values may change in the future.
Der.dev 🔥🛠️ retweeted
Pro tip to folks trying to do long horizon work with gpt 6.2 sol and Astra
If you’re seeing the model stop early, before trying goal mode, try increasing the reasoning effort.
While I don’t normally recommend working at max, if you want to let it work for hours, give it a shot
Der.dev 🔥🛠️ retweeted
One of the funny reasons for the difference in vibes between the influencers and the crowd today on 6.2 Sol/Astra is that we gave a few folks access a while ago and then made a lot of changes in the last 3-4 days and I'm pretty sure no re-testing was done within those days and the videos were produced last week.
Don't get me wrong, while the models are a hefty jump compared to their GPT predecessors, they are by no means perfect and there are a few things the teams are excited for the models to get better at in future iterations in terms of taste and niceness of use. Love all the feedback and we hear you!
Der.dev 🔥🛠️ retweeted
Although the new models are very smart, we have more work ahead of us on making the models nice to work with, and will prioritize that for upcoming releases.
But even if it takes a little extra work, the intelligence and capability are extraordinary.
Der.dev 🔥🛠️ retweeted
the negative astra reviews are simply a result of the model+harness being frustrating to work w/ and not that impressive (comparatively, it's still is super intelligent!), esp on the tail of fable 5.2 which is great.
reviews are consistent bcz experience has been consistent.
Der.dev 🔥🛠️ retweeted
On a positive note, it’s shocking how 6.2 Astra can spend five hours building a fully working feature with backend and frontend changes and provide an end-to-end tested video of it running successfully in a staging environment. The progress in the last six months is astonishing.
































