Alejandro AO 🤗

372 posts

Alejandro AO 🤗 banner
Alejandro AO 🤗

Alejandro AO 🤗

@_alejandroao

🤗 dev advocate @huggingface 🤖 i help you build ai agents with open models, just reach out

Paris Katılım Kasım 2022
272 Takip Edilen638 Takipçiler
Sabitlenmiş Tweet
Alejandro AO 🤗
Alejandro AO 🤗@_alejandroao·
Hugging Face is more than a model catalog. In my new video, I break it down into: 1) Models + Inference Providers 2) Datasets + Data Studio 3) Spaces + Arena demos If you are deciding where to start in the HF ecosystem, this gives you a clear map.
Alejandro AO 🤗 tweet media
English
2
8
23
1.6K
Alejandro AO 🤗
Alejandro AO 🤗@_alejandroao·
Hugging Face is more than a model catalog. In my new video, I break it down into: 1) Models + Inference Providers 2) Datasets + Data Studio 3) Spaces + Arena demos If you are deciding where to start in the HF ecosystem, this gives you a clear map.
Alejandro AO 🤗 tweet media
English
2
8
23
1.6K
Alejandro AO 🤗 retweetledi
Ben Burtenshaw
Ben Burtenshaw@ben_burtenshaw·
tomorrow we are hosting a workshop on agentic evaluations. agents are shipping faster than our ability to evaluate them. time to fix that.
Ben Burtenshaw tweet media
English
1
6
13
630
Alejandro AO 🤗
Alejandro AO 🤗@_alejandroao·
@gdb all the cancer researchers complaining about this cancer story sound a lot like senior SWEs complaining about AI coding last year
English
0
0
0
20
Alejandro AO 🤗
Alejandro AO 🤗@_alejandroao·
this is why gpt 5.2/5.3 codex feels so much superior to other LLMs at coding. in scientific tasks, it’s the most cautious model (almost never makes mistakes) and it’s the most self-aware about its own biases. this research is not getting enough attention. watch the video first, then read the paper. it’s pure gold.
Alejandro AO 🤗 tweet media
David Louapre@dlouapre

GPT 5.2 drains your time being overcautious. Grok 4 wastes resources being reckless. This is what we found benchmarking 16 AI models on Eleusis: a game where you form hypotheses, test them, and refine theories, just like real science 🧪♠️ New video on @huggingface channel ⬇️

English
0
1
8
626
Alejandro AO 🤗
Alejandro AO 🤗@_alejandroao·
@vSouthvPawv great point. subagents + hooks do increase the reliability. still, the non deterministic aspect of agents is definitely the hardest part for a traditional swe! exciting times ahead
English
0
0
1
6
sovthpaw
sovthpaw@vSouthvPawv·
@_alejandroao Deterministic behavior should be managed with hooks. Skills work best when they compliment the role of the Agent (*ahem* system prompt). A general chat agent is less likely to select and properly use a stock monitor skill than a dedicated financial subagent with the same Skill
English
1
0
1
47
Alejandro AO 🤗 retweetledi
Ben Burtenshaw
Ben Burtenshaw@ben_burtenshaw·
hot take: I think 2026 will *feel* like the year of the harness, but actually be the year of multi agent systems. i.e. the better harness + model gets at individual tasks, the more we need to coordinate them
English
0
2
7
629
Alejandro AO 🤗 retweetledi
Ben Burtenshaw
Ben Burtenshaw@ben_burtenshaw·
new video on @huggingface youtube where I deep dive into how the kernels community makes it easier to build and use optimized kernels. if you don't know, optimized kernels squeeze more out of hardware by adapting the operation with CUDA or C code. they can be complex to build and distribute, sometimes requiring hours to install. the kernels community takes this down to a few seconds. thanks to AIFoundry.org for hosting the event and making the video.
Ben Burtenshaw tweet media
English
2
5
38
1.5K
Alejandro AO 🤗
Alejandro AO 🤗@_alejandroao·
@plumbuns that’s because probability is nature-coded. calculous is human-coded.
English
0
0
10
1.4K
sarv
sarv@plumbuns·
how the FUCK does calculus make more sense than probability
English
277
1.4K
18K
416.8K

Keşfet

@gdb @ariG23498 @vSouthvPawv @huggingface @feelzyou @plumbuns @elonmusk @BarackObama