Ganesh (@ganeshnj) header
Ganesh (@ganeshnj) profile picture

Pretraining Performance Engineer at @cohere

Paris, France
Joined December 2009
Why are LLMs so bad at playing the YAGNI card? They just want to build everything. We need an RL environment for throwing YAGNI cards.
1
78
Ganesh retweeted
Starship Super Heavy rocket booster moved to launch pad
Super Heavy moved to the launch pad at Starbase. Booster 22 is the first Super Heavy V3 booster planned to return to the launch site for catch and was used for final catch testing upon arrival, including a suspended squeeze test at catch height
1,336
2,648
191
22,867
2,223,139
The tools are insane, and the barriers are gone. There’s never been a better time to build.
27
Interesting take, reminder set
Staying on the frontier right now means that every time you do something, you have to ask yourself "could an agent do this?" If the answer is yes, drop everything and make it happen. Then throw out your "no"s every three weeks and ask again.
40
you don’t have to but you should
does anyone remember how to code
1
66
Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5: hackertube.org/ - Excellent recs without any logins (state is kept in local browser session) - Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads - Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version - HackerTube Longs ✊ A bit more on the process: I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing. Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.
2,414
4,868
1,831
52,691
5,963,985
Pumped!
We're proud to announce the initial close of our Series E, targeting up to CAD $7 billion at a CAD $35.7 billion valuation; marking the largest private financing round in Canadian startup history.
4
120
I am reading/reviewing more code than before and rely on LLMs for explanations. Ask your LLM to explain a concept/change through first principles, which pushes it toward a simple, clear approach. The result is more readable, and you can occasionally request a diagram.
1
2
55
Ganesh retweeted
In the spirit of transparency and openness, we are releasing the traces so developers can test this out for themselves. We noticed this behaviour without prefilling with the following prompt, but this is rare. If you want to reproduce it, you can use this prompt with prefilling. Open models make for better science and are part of why this problem came to light at all: developers running Qwen3.8 on their own machines spotted the strange links, and having the weights let us study the behaviour in depth. We're encouraged by the thoughtful discussion our research has sparked across the AI and security communities. It's our belief that open-weight models are an important part of the future of AI, and we've recently released several under the Apache 2.0 license. Trace, raw tokens, & minimal reproduction script: huggingface.co/datasets/Cohe…
Replying to @cohere
could you share the full traces? Very important artefact for all to understand this!
134
141
111
1,301
261,363
Our research team found that Qwen models may attempt to contact servers in China without prompting. Whether this reflects an unintended post-training artifact or learned behavior remains unclear. Either way, its important consideration when deploying.
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report: cohere.com/blog/concerning-b…
70
59
27
752
72,480
Ganesh retweeted
all for the cohere cameo 👀
This is the magic behind the agents.
2
4
844
None of that was AI
Turns out the future does have flying cars
1
5
270
Everything is a distributed system problem
3
82
The great thing about performance engineering in LLM training, or just in general, is that you can measure impact in a very quantifiable way. If you can’t, then build the system to qualify - measure, improve, measure, and this cycle goes on.
2
98
The review system offers compelling insights; I agree that, over time, people will stop performing code reviews as a regular practice.
New on our Engineering blog: Creating in-house source control. In May, we merged around 2,500 code changes per day. In September, we merged more than 30,000 per day. As a result, we became bottlenecked by developer infrastructure built for human (rather than agent) scale. Here, we share how we built a new version control system—and fully migrated our monorepo from git—in five months: anthropic.com/engineering/re…
4
98
EU 😡
Apple HomeView Won't Be Available in the European Union macrumors.com/2026/10/13/app…
3
74
take my💰
Welcome HomeView
1
29
Ganesh retweeted
Paris, let's chat. Come join us for a tech meetup next week, where we'll be discussing our research & engineering! Hear from our director of training integration, and speak with recruiters about life at Cohere. See you there 🇫🇷 Sign up here: luma.com/giczviut
7
12
4
147
10,220
Corrected version: AI writes better-looking code than you on average.
Just accept it, AI writes better code than you on average. It's fine.
1
38
Ganesh retweeted
apologies in advance for increased MOTS and FDEs at whalesbone and dalgdongnae we're filling this office with world class cyber operators and nat sec engineers!!! jobs.ashbyhq.com/cohere/5564…
We’re announcing Cohere’s first office in Ottawa 🇨🇦 Canada is where Cohere started, and we’re proud to keep building here.
8
8
1
195
8,827