Maxime Voisin
@maximevoisin_ai
Followers
908
Following
1K
Media
32
Statuses
735
Product manager Tools/Agents @cohere. Previously @labelbox, @stanford computer vision labs. Cofounded Splash (failed AI x news startup).
Joined June 2020
Couldn't be more proud of the team for building one of the strongest Deep Research agents out there!! . Cohere Deep Research is coming for enterprises. fully private. deployable on-prem. on just one GPU ๐คฏ. Exciting times yiha.
Command A Reasoning powers end-to-end systems involving chained and hierarchical agents and leveraging the most relevant tools to accomplish tasks โ for example our Deep Research system, which is coming soon to North.
0
5
44
8/ Iโd build it, but Xโs API is too expensive for external product builders to build anything on top of it. @nikitabier @elonmusk.
0
0
1
7/ Basically for $5/user/month you can build the most hyper-personalized and hyper-relevant content machine ever. I'd happily pay $100/month for it. Surface the most relevant world knowledge for each mind. Connect humanity to insights that matter to them.
1
0
1
6/ It roughly costs you $5 per user per month: . Surface the top 5K threads/day (like you already do using X's algo) x 3 tweets per thread = 15K tweets. ~25 words/tweet = 375K words/day. 30 days โ 11.25M words, ~15M tokens. At $0.30/M tokens (Grok 3 Mini), that's $4.50/month.
1
0
0
5/ Why it matters: No guessing what user want in their feed - I say it. Businesses will pay thousands per month. Think monitoring competitors, product launches, industry updates, market research, R&D, comms managers. .
1
0
0
4/ Bonus: .- Also give me a daily summary for quick overview. - Let users drop detailed feedback to refine their daily curation, beyond likes/dislikes. - Batch processing and cached tokens keep costs low.
1
0
0
3/ How it works: Do some low-level search to surface the top ~5,000 daily tweets. Use X's algo for this. Then feed these tweets to Grok in order to score relevancy based on my hyper-specific description. Package me the top 50 posts, once a day or every few hours.
1
0
0
2/ Example hyper-personalized feed I'd pay for: "Surface research content discussing tool use and agentic architectures for LLMs. Favor tweets backed by public papers, not company claims. [ plus 2 more pages of instructions to make it hyper-personalized and hyper-relevant ]".
1
0
0
1/ Hey @nikitabier, X product idea. I'd happily pay $100/month for it. Let me provide a 3-page text or voice describing the content I care the most about, and then get the 50 most relevant posts every day. No endless scroll. It costs you $5/user/month (see end of thread).
2
0
2
RT @ArtificialAnlys: Cohereโs Command A API combines the intelligence leap from their prior models with a significant speed boost. https://โฆ.
0
2
0
Great day for enterprises: deploy the best AI directly into your infrastructure. Max performance. Max security. Minimal compute. Cohere just released the best on-prem AI model:.โก๏ธ Outperforms all open models. โก๏ธ Matches performance of bleeding models. โก๏ธ Fully air-gapped
1
3
45
and now the best open vision model. climbing up the various LLM fronts. congrats Cohere for AI team!.
LFGG! Cohere just dropped Aya Vision 32B and 8B - beats Llama 3.2 90B Vision and Gemini Flash ๐ฅ. > 8B Model: Best-in-class for its parameter size, outperforming competitors by up to 81% win rates . > 32B Model: Outperforms models 2x its size, achieving 49%-72% win rates. Some
0
1
16
some good weights ๐ฒ๐.
Introducing Command R7B: the smallest, fastest, and final model in our R series of enterprise-focused LLMs!. It delivers a powerful combination of state-of-the-art performance in its class and efficiency to lower the cost of building AI applications.
0
0
26
RT @cohere: Tool generations by Command models just got better! Weโve updated our Structured Outputs so that our models now strictly adhereโฆ.
0
12
0
RT @Nils_Reimers: ๐๐๐ฎ๐ง๐๐ก ๐จ๐ ๐๐จ๐ก๐๐ซ๐ ๐๐๐ซ๐๐ง๐ค ๐.๐ - ๐๐จ๐จ๐ฌ๐ญ ๐ฒ๐จ๐ฎ๐ซ ๐๐๐๐ซ๐๐ก ๐. What is new:.- Large gains in multilingual retrieval ๐บ๐ณ.- Reasoning Caโฆ.
0
19
0