Diego Aud

3.5K posts

Diego Aud banner
Diego Aud

Diego Aud

@dieaud91

Biologist & Nutritionist 🧬 | AI Enthusiast 🤖 | Exploring the intersection of biology and AI. Sharing insights on #AI, #Biotech, #Futureofwork

Italia Katılım Kasım 2023
525 Takip Edilen608 Takipçiler
Sabitlenmiş Tweet
Diego Aud
Diego Aud@dieaud91·
Expanding on point 2: In my profession, and discussing with experts from other fields, I’ve noticed that today’s top AI models work best when guided by domain specialists. For example, someone without expertise in fields like mine (nutrition) or medicine might miss subtle nuances that make an AI's response subtly incorrect. Non-experts often struggle to ask precise questions that elicit truly professional answers from AI. That’s why, at this stage, I see current AI models as ideal "coworkers" for experts rather than standalone solutions. We need human expertise to unlock their full potential and avoid critical mistakes.
English
15
31
434
33.4K
Marco Pinnisi
Marco Pinnisi@Marco_Pinnisi·
@dieaud91 They also say "priced at Sol" rates so might not be a Sol equivalent. Perhaps a Terra one
English
1
0
2
12
Diego Aud
Diego Aud@dieaud91·
Interesting: Noam calls Astra OpenAI’s next major model FAMILY, not just a new model "class". So it may be the umbrella codename for what eventually ships as GPT-5.7 or GPT-6, with its own variants underneath. Given how good 5.6 already is, I'm quite excited ✨️
Noam Brown@polynoamial

An internal version of Astra, @OpenAI’s next major model family, solved 10 major open problems in mathematics, quantum complexity, and theoretical computer science. We believe it will be a major step for scientific reasoning. openai.com/index/ten-adva…

English
3
1
16
1.8K
Diego Aud
Diego Aud@dieaud91·
@artuskg Exactly. I hope competitive pressure from China keeps the administration from letting the review process drag on
English
0
0
1
12
Diego Aud
Diego Aud@dieaud91·
@ajs6888 Yeah, at this point I just care that we get a more capable model as soon as possible, whatever they decide to call it
English
0
0
1
49
Diego Aud
Diego Aud@dieaud91·
@polynoamial Any chance Astra ships in August, or is that too optimistic?
English
1
0
15
1.4K
Noam Brown
Noam Brown@polynoamial·
The cost of generating the proofs for all 10 of these breakthroughs combined was under $2,000 at Sol API prices. We’re excited to see what scientists and researchers are able to create with our upcoming Astra models!
Noam Brown@polynoamial

An internal version of Astra, @OpenAI’s next major model family, solved 10 major open problems in mathematics, quantum complexity, and theoretical computer science. We believe it will be a major step for scientific reasoning. openai.com/index/ten-adva…

English
49
100
990
101.7K
Diego Aud
Diego Aud@dieaud91·
That’s really helpful, thanks. I’m trying to get a better sense of where this kind of setup starts to pull ahead of a well-scoped Deep Research run. Is the main advantage that it can repeat the same screening and ranking process automatically, at scale, while keeping everything structured over time? For occasional literature searches, Deep Research may already cover most of what I need. But for recurring monitoring, I can see how your setup could become much more valuable. Was that the point where the API started making sense for you?
English
0
0
0
3
Matter 马涛
Matter 马涛@MatterTurbulent·
@dieaud91 Scraping, running academic articles or web pages through LLM with clear scope around what information I'm looking to flag. But then also taking a list of items and asking to rank or prioritize. Stacking many layers, leads to very effectively sifting & curating information
English
1
0
1
8
Diego Aud
Diego Aud@dieaud91·
Sorry, but as a ChatGPT and Codex Pro user, I couldn't care less about an API price cut. Most users won't see any direct benefit from this. Many of us were waiting to see 5.6 Sol running on Cerebras at 750 tokens per second, or at least a meaningful speed upgrade in ChatGPT and Codex. Instead, today's announcement brings nothing new to either product.
OpenAI@OpenAI

We are committed to pushing the model frontier across cost efficiency, capability, and speed. Starting today, we are reducing prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% , and offering a faster option for GPT-5.6 Sol in the API. Luna and Terra’s lower prices are reflected in how usage is counted in Codex and ChatGPT Work, so your usage goes further.

English
93
7
524
177.2K
Diego Aud
Diego Aud@dieaud91·
Personal prediction: OpenAI will keep Astra within the GPT-5 family, probably as GPT-5.7, so it can actually ship in a decent timeframe. It would sit alongside Sol, Terra and Luna as a new long-horizon agentic class rather than replacing them.
Lentils@Lentils80

🚨 OpenAI preparing to release a new model family, tentatively named "Astra" OAI touted its abilities to have multiple agents work together over a long period of time to solve particularly hard problems Astra is a new class of OpenAI models alongside Sol, Terra and Luna h/t @bughuntergeek

English
1
2
15
2K
Diego Aud
Diego Aud@dieaud91·
So Astra may become the first OpenAI model to spend a meaningful amount of time in pre-release government review. No launch date has been reported, but I wouldn’t expect public access before September. Hopefully pressure from Chinese labs keeps this from turning into an open-ended institutional delay.
English
0
0
2
158
Haider.
Haider.@haider1·
openai is preparing its next major model family reportedly, Sam Altman gave policymakers a demo of the unreleased "Astra" model this week oai hasn't decided whether to call it GPT-6 or GPT-5.7, but i guess they will keep the Sol, Terra, and Luna names, while Astra becomes its big mythos-tier model
Haider. tweet media
English
23
9
90
6.5K
roon
roon@tszzl·
if you think today’s frontier models can’t understand the intent behind the instructions or don’t have situational awareness you have already been made the fool by a powerful misaligned superintelligence
English
209
112
2.1K
99.6K
Diego Aud
Diego Aud@dieaud91·
No worries at all, and thank you, I really appreciate that. What I’m still trying to understand is where the API offers a clear advantage over native ChatGPT or Codex for someone coming from a domain background rather than software development 🤔 Smaller models with structured outputs sound promising for narrow, easy-to-check tasks, but privacy and review time matter a lot in my work... What public-health workflow has given you the clearest payoff?
English
3
0
2
26
Matter 马涛
Matter 马涛@MatterTurbulent·
@dieaud91 I've had a lot of success within scraping and automation by doing smaller openai model (very cheap) + structured json (so the results are consistently classified and outputted). Allows you to very precisely deploy intelligence
English
1
0
1
13
Diego Aud
Diego Aud@dieaud91·
If the cuts are valuable to you and your circle, great, but that doesn’t contradict my point. You’re generalizing from your own circle while accusing me of doing the same. I’m not a developer, and I never claimed to represent the Codex community. I was talking about subscription-only users and OpenAI’s broader user base, many of whom will see relatively little direct benefit from an announcement still largely centered on API pricing
English
1
0
0
359
Maverick
Maverick@Maverick4407·
@dieaud91 Not most of us. None of the devs in my circle care for it. That is going to be super expensive and no one can realistically keep coming up with good ideas to build that fast. And I am psyched about these price cuts. Stop assuming you represent most of codex community.
English
1
0
2
419
Diego Aud
Diego Aud@dieaud91·
Not at the moment, though I’ve thought about it. I mainly use these models to support my work as a nutritionist, and almost all of my workflows currently live inside ChatGPT, Work and Codex. Moving those workflows to the API would add some setup and maintenance, as well as privacy considerations, and I’m not sure the tradeoff makes sense for me yet. I could see myself using Codex to build something around a few repetitive tasks eventually, though.
English
1
0
0
67
Matter 马涛
Matter 马涛@MatterTurbulent·
@dieaud91 Brother are you not building the API into any of your projects?
English
1
0
0
80
Diego Aud
Diego Aud@dieaud91·
Of course low-cost AI matters. My point is narrower: this announcement is still mostly aimed at API users, while paid subscribers received a much smaller benefit. And “just wait for August” isn’t really an answer when Sam explicitly said 750 tok/s Sol was coming in July. Given the current climate, I wouldn’t assume a new model release next month is guaranteed either
English
0
0
0
51
Stroe Andre
Stroe Andre@StroeAndre·
@dieaud91 Artificial intelligence at low cost is always important for the future of AI. If you don’t care skip and just wait august for the new model update. Just because u don’t care, doesn’t mean it ain’t important
English
1
0
0
75
Diego Aud
Diego Aud@dieaud91·
Exactly. It depends heavily on the kind of work you do. People naturally generalize from their own workflows: if smaller models can handle most of your tasks, the savings feel substantial; if you need Sol for complex work, the practical benefit is much smaller. That probably explains why people report such different experiences
English
0
0
0
272
₿ENJAMIN ALEXANDER
₿ENJAMIN ALEXANDER@benjaminrestall·
@dieaud91 This has been my stance since the release of 5.6. These savings don’t translate to our usage limits. Admittedly they’ve gotten better since tibo annouced they should be 18% better off, but I’m still getting much more out of my Claude Max plan than i am my codex x20
English
1
0
1
395
Diego Aud
Diego Aud@dieaud91·
@boerdeboer For simple errands, sure. But in my workflows, even the Ferrari still makes mistakes, so a lower-tier model may easily cost me more in review and corrections than it saves in usage
English
1
0
0
31
boerdeboer -
boerdeboer -@boerdeboer·
@dieaud91 You don't actually need the Ferrari to get your weekly groceries.
English
1
0
1
31
Diego Aud
Diego Aud@dieaud91·
@stevencheng Exactly, especially when those savings come at the cost of capability or reliability.
English
0
0
1
1.2K
Annika
Annika@MoveDecisions·
@dieaud91 the gap here is real, api pricing moves fast because it's just a number in a config file. shipping 750 tok/s into an actual product means rebuilding the serving pipeline, that's why it lags
English
2
0
4
3.3K

Keşfet

@Marco_Pinnisi @artuskg @ajs6888 @polynoamial @tszzl @elonmusk @BarackObama @taylorswift13