Pol Avec
@pol_avec
Followers
502
Following
3K
Media
188
Statuses
2K
鵬仁。Developing ML for fun. And enjoying the wild ride @fewsats. AI, space, and the transcend. Zima Blue
Barcelona
Joined March 2018
AI is the new UI: The future of interfaces is generative. @answerdotai 's FastHTML + @htmx_org are uniquely positioned to unlock truly interactive AI experiences beyond text chats. 150 lines of pure Python can create fully interactive AI apps like these. Let's see how 👇
4
10
27
"reform publish-or-perish incentives". yes please.
The mainstream view of AI for science says AI will rapidly accelerate science, and that we're on track to cure cancer, double the human lifespan, colonize space, and achieve a century of progress in the next decade. In a new AI Snake Oil essay, @random_walker and I argue that
0
0
0
🤖💲.
𝐀𝐠𝐞𝐧𝐭𝐢𝐜 𝐜𝐨𝐦𝐦𝐞𝐫𝐜𝐞 𝐢𝐬 𝐞𝐱𝐩𝐥𝐨𝐝𝐢𝐧𝐠 🤖 💸 . From AI concierges to shopping copilots, new companies are owning the customer experience but hitting a wall at checkout. Most stores don’t expose APIs. Scraping is easy. Buying isn’t. And payments? Even harder.
0
0
0
👀.
am i hallucinating or did we just go from "openai acquiring windsurf" to "cog+ws team shipping wave 11" in seven days.
0
0
0
"ChatGPT mimics the intimacy that people desperately want.". For people that didn't know before, please have a look. For those of you already aware, please take the time to explain to someone who doesn't. A small effort on your side, could make someone's life better.
Psychiatrists have been warning about the potential for chatbots to trigger psychosis for some years.
0
0
0
humans are quite bad dealing intuitively with probabilities. more on this topic in the interesting book "Fooled by Randomness".
A problem every AI company is wrestling with. Why is testing LLMs so frustrating for developers?. Because even a perfect release won’t hit 100%.You’re not writing unit tests, you’re managing probabilities. 🤖🎙️From my latest Agents at Work interview with @KorNimrod (CTO Baz)
1
0
1
don't let reality spoil a good narrative.
It’s funny that people on this site think major LLM efforts are talent-bound rather than org-bound. The talent differential has never been big between major orgs. Most of the difference in outcomes is due to organisational factors - like allocating compute to the right bets, and.
0
0
2
aggregating statistics is often an illusion of progress.
I always learn a lot more from in-depth analysis of few random cases over dashboards of aggregate statistics across all cases. Both projections can be helpful but the latter is disproportionately pervasive.
0
0
1
I didn't see this one coming.
Cognition has signed a definitive agreement to acquire Windsurf. The acquisition includes Windsurf’s IP, product, trademark and brand, and strong business. Above all, it includes Windsurf’s world-class people, whom we’re privileged to welcome to our team. We are also honoring
0
0
0
RT @BTC_Archive: 💥Bitcoin's Global Takeover w Autonomous Ai Agents . I talk to Jordi Montes (@positiveblue2) about:.- Bitcoin micro-payment….
0
118
0
Very interesting outside view into RL Atari from John. You should watch the talk as he's very clear and articulate, so I won't repeat his points. Having dabbled briefly with RL for fun, it's good to see how many issues he uncovers just by adding the twist of playing the game.
0
0
1
a very underrated feature of LLMs is how little data they consume (just text), great for working in slow networks like airport or limited data roaming.
0
0
0
I think telling Claude to reflect on how to improve its prompt is one of the best ways to improve Claude Code. e.g. "is there any lessons learned we should persist to your prompt before we begin working on X?". it will:.- help you prevent same issues from happening again.-.
1
0
1
AND I just discovered CC is now included in the monthly plans (with some limits) instead of having to pay per token (which is quite expensive). You have to logout & login again.
ok, update due. After 10 days of Gemini CLI, I went back to Claude Code and there's no comparison. I very much prefer CC coding style and most of the proposals it sends my way. The difference is subtle, but after going back I do real feel the difference personally. I'd rather.
0
0
2
ok, update due. After 10 days of Gemini CLI, I went back to Claude Code and there's no comparison. I very much prefer CC coding style and most of the proposals it sends my way. The difference is subtle, but after going back I do real feel the difference personally. I'd rather.
Gemini CLI is pretty good. In general I prefer Sonnet coding style, but the price difference is significant enough (gemini is mostly free rn). afaik Claude Code has more features & MCP, but I haven't used them much so far. Never underestimate the power of FREE 😂👇
0
0
2
no framework will give you more features with less code 🔥. worth your time.
Did you know we have a complete, heavily commented, updated idiomatic FastHTML app that shows how to create a login system, drag/drop, all CRUD endpoints, etc in a single small Python app?. Read it yourself, or give it to your LLM as context. 😁.
0
0
1
