Diego Aud

3.2K 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 加入时间 Kasım 2023
461 关注525 粉丝
置顶推文
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
16
30
441
32.6K
Derya Unutmaz, MD
Derya Unutmaz, MD@DeryaTR_·
Stand by for another momentous day in the age of AI!
English
11
10
193
10.5K
Diego Aud
Diego Aud@dieaud91·
Just finished watching the livestream. I wasn’t hyped at first - I was hoping for 5.5 - but I have to admit this image-gen model looks seriously impressive. Feels like 2026 is the year image generation gets solved, especially if v3 drops before the end of the year...
OpenAI@OpenAI

Made with ChatGPT Images 2.0

English
0
0
6
247
Diego Aud
Diego Aud@dieaud91·
@flowersslop "Now includes tools to get researchers to spill the beans while hyping unreleased models"
Diego Aud tweet media
English
0
0
1
170
Flowers ☾
Flowers ☾@flowersslop·
my package finally arrived
Flowers ☾ tweet media
English
59
25
1.2K
53.7K
Diego Aud
Diego Aud@dieaud91·
The compute desert is real.
Diego Aud@dieaud91

@AcerFur So they ditched the best checkpoint. Sad that we still have to optimize around budget constraints... The compute desert is real.

English
0
0
3
54
Diego Aud
Diego Aud@dieaud91·
@AcerFur So they ditched the best checkpoint. Sad that we still have to optimize around budget constraints... The compute desert is real.
English
0
0
5
447
Logan Kilpatrick
Logan Kilpatrick@OfficialLoganK·
Introducing our biggest upgrades to the Deep Research API yet... including Deep Research Max (our SOTA system), MCP support, Native charts & infographics, planning mode, full tool support (including Google tools), full multi-modal input support, & real-time progress streaming!
Logan Kilpatrick tweet media
English
118
143
1.8K
107.3K
Diego Aud
Diego Aud@dieaud91·
We’ve reached the inflection point. AI images are now good and accurate enough to be useful not just for day-to-day use, but for professional work as well; maybe not in the art world yet, but in many other fields, absolutely.
Mark Kretschmann@mark_k

GPT-Image-2 is here! 👌 The new image model is especially good with text rendering, as you can see here. It's rolling out right now to all OpenAI users, and should become available to you *today*. In fact you might already have it! Check this out:

English
0
0
20
1K
Diego Aud
Diego Aud@dieaud91·
@AndrewGinns @kfountou @OpenAI Sure. I just tried, but it looks like I can’t DM you first. If you send me a quick message, I’ll reply with the prompt and screenshots of the outputs
English
1
0
0
13
Diego Aud
Diego Aud@dieaud91·
@DrBeavisAI Yeah, it somehow feels 5.4-ish. Somewhat better, but not what I expected to feel from 5.5 Pro. I might be wrong. Hope that with Thinking they fix the tendency to "not think enough" even when max Thinking effort is selected. Very unsure about 5.5 on Thursday...
English
0
0
3
142
G, MD
G, MD@DrBeavisAI·
@dieaud91 I think it's still 5.4, tweaked a little. Somehow 3x faster. We need now faster 5.4 thinking too ! But, see post below. No 5.5 on Thur? x.com/DrBeavisAI/sta…
G, MD@DrBeavisAI

@chatgpt21 Also if they upgraded GPT-5.4 Pro, does that mean no 5.5 this week? Was the same when they made 5.2.chat better in Feb and took a month for 5.3/5.4 to arrive... Now 5.4 think needs same upgrades as Pro..

English
1
0
1
358
Diego Aud
Diego Aud@dieaud91·
New theory about "GPT-5.5": imagine if the allegedly shadow-dropped GPT-5.4 Pro we’ve all been using lately isn’t Spud itself, but a new version of GPT-5.4 Pro improved by Spud behind the scenes 💀
English
2
2
56
3.3K
Diego Aud
Diego Aud@dieaud91·
@jinayx Really hope so. It'd be awesome as the real 5.5 Pro would be even more precise/sharp
English
0
0
1
20
Jinay Shah
Jinay Shah@jinayx·
@dieaud91 I'm with you as well, the model running under 5.4 Pro looks like normal 5.5 thinking
English
1
0
1
13
Diego Aud
Diego Aud@dieaud91·
It's a very long and complex prompt (nutrition task, I'm a biologist). The previous model had roughly the same amount of formatting errors (bold, spacing) or a bit less, but usually got the calculations right based on what was written. ​Yesterday, I noticed a table where the calculations were all off by 30-50 kcals. If you want, I can share it via DM. However, to be transparent, I've only seen this specific error happen once, and my sample size for both this and the previous model isn't huge either. I've only been experimenting with this task for 2 weeks at most, so it could also just be a fluke. What I can tell is that - for this specific task, it doesn’t feel like a perceivable upgrade
English
1
0
0
74
Andrew Ginns
Andrew Ginns@AndrewGinns·
@dieaud91 @kfountou @OpenAI Thanks, and this is a task that was previously successful? If you can share the conversation (or any others that demonstrate where it's lacking) that would be ideal!
English
1
0
1
114
Diego Aud
Diego Aud@dieaud91·
@tmfpretty Yeah, that's what I'm starting to consider as well
English
0
0
2
202
Jeremy Pretty
Jeremy Pretty@tmfpretty·
@dieaud91 Spud has to be 6.0 ... 5.5 is just a sampling of whats to come
English
1
0
2
242
Diego Aud
Diego Aud@dieaud91·
Personally, I've noticed a few occasions where the model seemed a bit less adherent to my instructions (a very long set). In my case, the task involves generating a .docx file with specific rules and styling, while simultaneously performing specific calculations to populate the tables
English
1
0
0
116
Diego Aud
Diego Aud@dieaud91·
@antonpme @MoeCanDoIt I think that AI labs should give users an approximate size of the models, as xAI is doing. I want to know if I'm interacting with a 500B or with a 1.5T parameter model, so I know what to roughly expect. The big model smell is real.
English
1
0
2
31
Anton P. 👽
Anton P. 👽@antonpme·
@MoeCanDoIt As I tweeted today, Opus 4.7 feels like Sonnet 4.7, not Opus at all... it lacks that EQ that Opus always had.
English
1
0
4
159
Moe
Moe@MoeCanDoIt·
Opus 4.7 is somehow DUMBER than 4.6. How do you regress a model between versions?
English
46
6
199
8.1K
Diego Aud
Diego Aud@dieaud91·
@0xRizzler Good point. I still hope there's more room (for my tasks at least) with the released 5.5 pro model
English
1
0
0
21
🥀
🥀@0xRizzler·
@dieaud91 if it was just more GPUs, the thinking pattern shouldn’t change this much speed yes, behavior no
English
1
0
0
16
🥀
🥀@0xRizzler·
GPT 5.4 Pro Extended went from 30 minute thinking down to 5 Something changed under the hood and they are not telling us Either they found a shortcut or a new model is being tested Anyone else noticing this or am I going crazy
Diego Aud@dieaud91

10th time in a row that GPT-5.4 Pro Extended thinks for just 1 to 5 minutes instead of the typical 15-30 minutes. Same kind of prompt. ​Right before this response, I got an "evaluate this answer" popup, but it disappeared due to an error. I strongly suspect they're testing a new model or tweaking something under the hood. ​Has anyone else noticed this drop in thinking time?

English
1
0
7
281
Diego Aud
Diego Aud@dieaud91·
@gonzalo_bruna @bughuntergeek Hope it's 5.5 Thinking and not 5.5 Pro. It's good but, for my specific tasks, I don't notice much difference in terms of precision/complex instruction following (it generated a docx with roughly the same number of small mistakes as 5.4-Pro)
English
1
0
2
39

发现

@DeryaTR_ @flowersslop @AcerFur @OfficialLoganK @GoogleAIStudio @AndrewGinns @kfountou @OpenAI