Principal Engineer at Rippling
San Francisco, CA
Joined August 2011
- Tweets183
- Following3.3K
- Followers300
- Likes1.9K
Ajit Singh retweeted
BREAKING: GitHub to announce a breakthrough eight eights of availability, for eligible customers.
Ajit Singh retweeted
The GPU market right now is pure financial engineering.
You can't get them on demand. So you sign a 3-year contract. But to get that contract, you need to put 30% down upfront.
Most companies can't do that. So they find someone who can.
- that person puts down your 30%
- you sign the 3 year contract
- your payments cover the entire cost of the GPU plus some
- the person who put down your deposit owns the gpu outright
So basically they took on zero hardware risk, got paid back entirely through your contract, and even kept the asset at the end.
It's not a chip shortage anymore. Everywhere I see it's a capital access problem.
If you have the 30% money, you're literally printing it right now. If you don't, you can't even get in the door.
Ajit Singh retweeted
I was always slow on Forward v.s. Reverse KL. Then I realized a mnemonic trick.
FOrward -> FOMO
Fear of missing out on some probability mass:
min_q KL(p || q) encourages mode coverage of p.
Reverse -> Retreat
Retreat to a high-density mode:
min_q KL(q || p) is mode seeking.
Ajit Singh retweeted
Easy to remember:
Video model: f: S ➔ S
World model: f: S × A ➔ S
Ajit Singh retweeted
what changed is that evals are now RL training data
apropos of the LM arena raise — I'm skeptical of eval startups, because so many failed in previous waves.
why?
1) great eval ppl should probably work on post-training instead
2) customers graduate quickly to doing their own evals
3) labs try rly hard to goodhart eval startups
ALT Why eval startups fail May 8th, 2025 Why are there so few independent eval startups? Whenever there's a new AI trend, like agents, or voice, or voice agents, developers are faced with a flurry of options, and a subset of them are convinced that there's a business opportunity in identifying the best models and selling that knowledge to other developers—that is, selling evals. I've seen this in every wave of generative AI, since before we were calling it generative AI. I haven't seen any succeed, outside the safety evals niche. I have a few theories why independent eval startups die. First, people who can design and run good evals can make more money and have more influence in other parts of the model development stack, so talent attrits. Second, eval startups have a hard time finding customers, because clients have to be technical developers who want to build with APIs, but also not technical enough to run their own evals. And third, eval startups face immense optimization pressure t
Ajit Singh retweeted
I’m stoked to share our new paper: “Harnessing the Universal Geometry of Embeddings” with @jxmnop, Collin Zhang, and @shmatikov.
We present the first method to translate text embeddings across different spaces without any paired data or encoders.
Here's why we're excited: 🧵👇🏾
Ajit Singh retweeted
It's deeply concerning that one of the best AI researchers I've worked with, @kaicathyc, was denied a U.S. green card today. A Canadian who's lived and contributed here for 12 years now has to leave. We’re risking America’s AI leadership when we turn away talent like this.
Ajit Singh retweeted
Starting salaries at Cambridge are less than $70k for most fields. Stanford? Double that.
As someone trying to hire the best global talent to work on UK and EU projects, I can't compete with the American universities. One election won't change that.
Ajit Singh retweeted
We should call models like Llama 3, Mixtral, etc. “open-weight models”, not “open-source models”. For a model to be open-source, the code and training data need to be public (good examples: GPT-J, OLMo, RedPajama, StarCoder, K2, etc.). Weights are like an exe file, which would be ridiculous to call open-source.
Ajit Singh retweeted
This would essentially make it illegal for a foreigner to do a postdoc in the UK.
🚨 BREAKING: Rishi Sunak will increase the salary threshold so that EVERY migrant has to earn over £38k before coming to Britain in a major plan to slash migration
He will also announce new limits on them bringing family
[@JackElsom]
Ajit Singh retweeted
We remain committed to our partnership with OpenAI and have confidence in our product roadmap, our ability to continue to innovate with everything we announced at Microsoft Ignite, and in continuing to support our customers and partners. We look forward to getting to know Emmett Shear and OAI's new leadership team and working with them. And we’re extremely excited to share the news that Sam Altman and Greg Brockman, together with colleagues, will be joining Microsoft to lead a new advanced AI research team. We look forward to moving quickly to provide them with the resources needed for their success.
Ajit Singh retweeted
Launching today: AI-powered candidate feedback summaries in Rippling Recruiting 🚀
This new feature auto-synthesizes interview feedback into a digestible, helpful snapshot of each job candidate.
Stay tuned—more AI-driven features are coming soon🛠️
rippling.com/blog/bringing-a…
Ajit Singh retweeted
We investigate what features of tabular data explain the difference, for this, we modify tabular data to narrow the gap.
Smoothing the outcome in feature space narrows the gap: deep architectures struggle with irregular patterns. Tree models do not care about smoothness
6/9
Ajit Singh retweeted
What are the latest research trends in AI?
Explore all NeurIPS submissions from 1987 to 2022 in Atlas.
atlas.nomic.ai/map/neurips
Ajit Singh retweeted
A lot has been speculated about TikTok's recommendations. This is the first paper I've read by the team, and it has many interesting details: expirable embeddings, parameter server, online training... Good #recsys stuff arxiv.org/abs/2209.07663
