yummy tokens yummy
sunny madra@sundeep
“If your $500K engineer isn’t burning at least $250K in tokens, something is wrong.”
English
Christophe
5.8K posts
Christophe
@_Blef
open-source agentic analytics https://t.co/DgLno6RK6b
Paris and Berlin Присоединился Haziran 2009
235 Подписки676 Подписчики
gpt-5.4 has been overfitted to run tool calls or what, this is so strange
English
@DFintelligence le mec est sur windows depuis des siècles, qui a piraté le compte d'Anis?
Français
@joffrey Après, pour bosser sur ce sujet d'analyse spécifiquement, mettre 20k rows en contexte c'est de toutes façons pas la bonne chose à faire, l'analyse sera rarement bonne sans écriture de code d'analyse, imho.
Français
Je réitère l'importance de pouvoir connaître le contexte utilisé / restant, même dans les apps grand public.
Venant de faire analyser un CSV de 20 000 lignes à Gemini et ChatGPT, ça n'a pas loupé : une fois la limite de tokens dépassée, les hallucinations commencent et la conversation est bonne à jeter.
Et je ne parle même pas de Claude, qui dès 16k tokens perd déjà 21,8 % de ses capacités.
Joffrey@joffrey
OpenAI, Google, xAI, Anthropic: please add a context-window indicator to your apps. I’ve seen people ask for translations, recipes, and dozens of unrelated things in the same infinite chat. Why not. But I’ve also seen vibe coders use a single chat for an entire app, then wonder why the model “forgot” what it did a few days earlier, probably 10M tokens ago. Branching is also mandatory now (thanks, ChatGPT!). How frustrating it is to accidentally pollute a conversation, only to realize that the only way to stop the damage is to go back and edit an older message—or else trigger hazardous compaction, or blindly deal with hallucinations and memory loss. We need to know when it’s almost time to start over, and we need a way to continue from a clean point without losing information. Please.
Français
Christophe ретвитнул
Feels pretty good to use the very product you built to create new features and do data analysis. All in the same place, all in nao.
English
Christophe ретвитнул
I've tried Cursor, Windsurf, Kiro, Warp, Claude Code, etc.
They all have session-based chats but fail to provide a centralized and persistent agent whose knowledge and performance improve over time.
Crazy how the chat paradigm has stuck.
When will we break the cycle?
English
Christophe ретвитнул
🚀 Excited to introduce Qwen-Image-Edit!
Built on 20B Qwen-Image, it brings precise bilingual text editing (Chinese & English) while preserving style, and supports both semantic and appearance-level editing.
✨ Key Features
✅ Accurate text editing with bilingual support
✅ High-level semantic editing (e.g. object rotation, IP creation)
✅ Low-level appearance editing (e.g. addition/delete/insert)
Try it now: chat.qwen.ai/?inputFeature=…
Hugging Face: huggingface.co/Qwen/Qwen-Imag…
ModelScope: modelscope.cn/models/Qwen/Qw…
Blog: qwenlm.github.io/blog/qwen-imag…
Github: github.com/QwenLM/Qwen-Im…
API: alibabacloud.com/help/en/model-…
English
had a quick look at the code, this is huge and there are interesting things to see regarding how next edit suggestions is developed
Visual Studio Code@code
VS Code: the open source AI editor. Starting today.
English
Christophe ретвитнул
Today we're releasing FLUX.1 Kontext - a suite of generative flow matching models that allow you to generate and edit images.
Unlike traditional text-to-image models, Kontext understands both text AND images as input, enabling true in-context generation and editing.
English
Christophe ретвитнул
Read the full post to see commentary from data leaders in the field - @TobiM, @royhasson, @_Blef - who give different perspectives of the current ecosystem on the managed Iceberg ecosystem, particularly AWS S3 Tables and Cloudflare R2 Data Catalog.
#commentary-of-data-engineers-in-the-field" target="_blank" rel="nofollow noopener">rilldata.com/blog/the-open-…
English
It's just about time all the MCP out there get this kind of vuln
invariantlabs.ai/blog/mcp-githu…
English
@ananthdurai 1. in the manifesto they say it's Iceberg compatible (catalog+data)
"Compatibility: The data and (positional) deletion files that DuckLake writes to storage are fully compatible with Apache Iceberg allowing for metadata-only migrations."
English
The DuckLake theme seems to speed up the metadata lookup. A couple of things I'm not sure
1. Does this still work with the Iceberg format? Do people need a migration from Iceberg?
2. How long will this design hold with AWS Express One slashing costs?
duckdb.org/2025/05/27/duc…
English
LinkedIn bots (created by LinkedIn team) are coming, this person does not even exist.
The funny part is that they do not even personalise the message "Hello LinkedIn Member".
English
As everyone, we also have Claude Sonnet 4 in nao, and we are testing Opus 4 to see how we can release it ✨
My first impression: this is way better than 3.7, which I completely skipped and I stayed on 3.5 since then.
English
@Fabien_Devos @qhoang09 @vybe_build @ycombinator This is an interesting look on the future, where every employee can build an app customized for his own context super fast
English
A year ago, I started working on a tool to build Web apps. But nowadays you can just chat with an AI and get the exact app you want. So I teamed up with @qhoang09 and we're super excited to announce @vybe_build, an AI to build internal apps, backed by @ycombinator !
English
Huge warm welcome to Claire and Christophe @_Blef from nao Labs (getnao.io), vscode fork agent for working with data, YC funded too recently :)
Pretty exciting to see more competitors in this space!
I'm working on helix.app.nz an AI Chart/dashboard assistant, nice to meet yall!
Welcome to building actionable business insights software supercharged with AI agents :D
English