kiratp

4 ms·

31.
by kiratp 2y ago
You walk into your nearest Chase (or similar) branch. Out in 10 mins.
32.
by kiratp 2y ago
I would urge you to not think this way: https://www.osmos.io/fabric
33.
by kiratp 2y ago
https://www.osmos.io/fabric Practical, real-world application.
34.
by kiratp 2y ago
Things have been moving so fast that it’s honestly hard for a small team to do that in parallel. I got to present at GCP Next about a part of this last year: https://www.youtube.com/watch?v=5QsM1K9ahtw I’m presenting in one
35.
by kiratp 2y ago
We use multiple post-trained models in production, at scale at https://osmos.io
36.
by kiratp 2y ago
The entire wrapped package of tested prompts, context management etc. is a whole step change from what you can build yourself. There is a reason Cursor is the fastest startup to $100M in revenue, ever.
37.
by kiratp 2y ago
You’re not using the best tools. Claude Code, Cline, Cursor… all of them with Claude 3.7.
38.
by kiratp 2y ago
Our EU customers use our technology to deal with all the invoices etc. they get sent as PDFs.
39.
by kiratp 2y ago
I hope this amount of effort means multi-GPU support is coming soon (saw you mention “this week” :) ) so we can tune QwQ on a single node!
40.
by kiratp 2y ago
Is my understanding here correct? Could this be the reason? https://news.ycombinator.com/item?id=43287312
41.
by kiratp 2y ago
Unless I’m missing something this isn’t online RL. They are collecting outputs in one pass and then doing a separate offline GRPO training run on those. The results of this paper would indicate doing what they did, but online could return b
42.
by kiratp 2y ago
It hilarious that the academic sector 1. publishes as PDF 2. spends all this energy on how to extract that info back from PDF 3. publishes that research as PDF as well. Receipt scanning is a multiple orders of magnitude more valuable busin
43.
by kiratp 2y ago
This isn't even close to true. Source: We have large EU customers.
44.
by kiratp 2y ago
It's shocking how much our industry fails to see past its own nose. Not a single example on that page is a Purchase Order, Invoice etc. Not a single example shown is relevant to industry at scale.
45.
by kiratp 2y ago
Nobody is using FP32 for AI. INT8, INT4, FP8 and soon FP4
46.
by kiratp 2y ago
Now compare the FLOPs
47.
by kiratp 2y ago
10K doesn’t get you 512 GB of VRAM in Nvidia land.
48.
by kiratp 2y ago
This is a fundamental misunderstanding of what a moat is in this industry. If GPT 4.5 had nothing of value to offer other than a flex of opening eyes ability to scale, it’s still a signal that they have the chops to throw the most amount of
49.
by kiratp 2y ago
https://www.telegraph.co.uk/news/2024/10/25/tommy-robinson-c...
50.
by kiratp 2y ago
> And since their share of the fruits of that invention was mostly bloodshed, job loss, and shattering of the world order they knew, I wouldn't blame them from being pissed off about getting the short end of the stick, and perhaps l
51.
by kiratp 2y ago
https://huggingface.co/docs/trl/main/en/gkd_trainer Take a set of prompts, run it through the large model and small model, calculate KL divergence between all the logits, then update the small model to m
52.
by kiratp 2y ago
Quantized floating point math can, under certain scenarios, be non-associative. When you combine that fact with being part of a diverse batch of requests over an MoE model, outputs are non-deterministic.
53.
by kiratp 2y ago
> just that GST is not usually shown until checkout unlike domestic sites. This still puts domestic sites at a significant disadvantage.
54.
by kiratp 2y ago
This is really dumb. Deepseek showing that you can do pure online RL for LLMs means we now have a clear path to just keep throwing more compute at the problem! If anything we made the whole "we are hitting a data wall" problem eve
55.
by kiratp 2y ago
r1-zero is pure RL with no SFT.
56.
by kiratp 2y ago
> I wonder if there are any AI tools out there doing context + planning + history + RAG + test + evaluation + tool use + feedback approaches. I've seen various tools do 2-3 of those. It's not a stretch to think that we might ev
57.
by kiratp 2y ago
> And if those policies should turn out to be too strict, and a few social media companies should find it no longer profitable to operate in the United Kingdom, that is not all that much of an issue for His Majesty's Government. The
58.
by kiratp 2y ago
Specifically regarding credit card security, it’s not really my problem is it? It’s the bank’s problem. The law makes it so.
59.
by kiratp 2y ago
Nvidia’s success is due to a decade+ of investment into their software stack (CUDA and friends), built on a world view that more software can and should be massively parallel. They were a key cause for LLMs being a thing in the first place.
60.
by kiratp 2y ago
Try sending the bob straight to an LLM and have it give you markdown. Bet it works.

More ›