Piotr Miłkowski

39 posts

Piotr Miłkowski

Piotr Miłkowski

@xeonbloomfield

Senior AI System Engineer at @callstackio -- Data Scientist 💻, PhD in AI 🔬 & petrolhead 🚗🏎️🏍️

Poland Katılım Şubat 2011
100 Takip Edilen36 Takipçiler
Damien Redecki
Damien Redecki@madebydamien·
→ looking for brand & marketing designer @callstackio → work hand in hand with me + @grabbou → AI, marketing, web, events, products, and everything in between → DM if this sounds like your thing → #job-2593561" target="_blank" rel="nofollow noopener">callstack.com/careers#job-25…
English
2
0
5
1.1K
Piotr Miłkowski retweetledi
Michał Pierzchała
Michał Pierzchała@thymikee·
In 𝚊𝚐𝚎𝚗𝚝-𝚍𝚎𝚟𝚒𝚌𝚎 v0.14 we completely reworked our skills: simpler instructions through 𝚑𝚎𝚕𝚙 command. 
But how did we know it’s better? We added tests with skillgym. 
Started at 41.7–54.2% passing on cheap models, and Codex got it to 100% while adding more cases.
Michał Pierzchała tweet media
English
4
4
53
5.7K
Piotr Miłkowski
Piotr Miłkowski@xeonbloomfield·
@retrovesto @thymikee Digging into the exact results of those two: Composer 2 managed to meet 2623/2730 requirements, while Composer 2 Fast passed 2590/2730 requirements - considering average of 10 runs. That's just a difference of 33 checks, which is 1.26% off the original score of 2623. Close! 2/2
English
0
0
1
52
Piotr Miłkowski
Piotr Miłkowski@xeonbloomfield·
@retrovesto @thymikee In theory they are, but in practice for GenAI you have to take into account that in order to retrieve the very same results you have to execute queries twice using the same seeds, over the same model instance, served by the same inference server used before. 1/2
English
1
0
1
23
Piotr Miłkowski retweetledi
Mike
Mike@grabbou·
OG team leaving Sopot. Damn, three days locked in hacking with @OpenMercato team or should I say - chatting with Codex 🤡 First time we got to enjoy the city! Shipping some cool stuff later next week! 🥵
Mike tweet mediaMike tweet mediaMike tweet media
English
0
3
24
2.4K
Piotr Miłkowski
Piotr Miłkowski@xeonbloomfield·
Self-driving companies won't run on prompts alone. They need systems that run work. @MulticaAI agents on @OpenMercato: - open-source, programmable CRM/ERP core - built-in permissions, workflows, APIs - human handoff when needed Building blocks for self-driving orgs.
English
1
5
13
2.2K
Piotr Miłkowski retweetledi
Mike
Mike@grabbou·
We're releasing Codex plugins for React Native development. Building, testing and device automation - all packaged together, with suggested prompts to get started. Try it out! ⬇️
Mike tweet media
English
8
27
305
21.2K
Piotr Miłkowski retweetledi
Michał Pierzchała
Michał Pierzchała@thymikee·
gpt-5.4 achieves better results than gpt-5.3 with 3,5x less tokens 🤯
Michał Pierzchała tweet media
English
17
11
263
28.7K
Piotr Miłkowski retweetledi
Callstack Engineers
Callstack Engineers@callstackio·
Which AI coding model actually writes good React Native code? We couldn't prove it either, so we built a benchmark. React Native Evals: 43 real-world tasks across animations, async state, and navigation. Read more on our blog ⬇️
Callstack Engineers tweet media
English
5
15
76
27.9K
NVIDIA GeForce
NVIDIA GeForce@NVIDIAGeForce·
We’re giving you TWO ways to WIN a one-of-a-kind GeForce RTX 4080 SUPER signed by NVIDIA CEO, and founder, Jensen Huang 👀 If you’re at CES head to our partner booths to enter 👉 nvidia.com/en-us/geforce/… Want to WIN here on social? ⚫Comment #RTXSUPER ⚫Like this post
NVIDIA GeForce tweet media
English
41.8K
3.6K
45.1K
1.8M
Piotr Miłkowski
Piotr Miłkowski@xeonbloomfield·
Led Zeppelin - Celebration Day... Something wonderful! ;-)
Piotr Miłkowski tweet media
English
0
0
3
0