HNHacker News
TopNewBestAskShowJobs

sarjann

157 karma · joined November 11, 2023

submissionscomments
sarjann··on Launch HN: Tokenless (YC S26) – Automatic model switching to save money
Sounds like this would be expensive on the input token side with some saving on the decode side. For workloads with large documents / context that could be an issue.
sarjann··on Launch HN: Tokenless (YC S26) – Automatic model switching to save money
If it only switches when cache is cold, it seems like it'd lock in one model for the whole session?

What use case are you imagining where cache eviction would happen during it? It might also be the case that a task might seem easy but then it ends up becoming hard for certain parts. For spawning subagents I could see this being useful maybe.

sarjann··on London Gatwick has launched a robotic airport parking service
I guess the argument might be that it'll be faster than finding a spot. Slightly minor benefit is you don't need to remember where it's parked.
sarjann··on Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber
Unfortunately them giving up coding means they have less traces to train on.
sarjann··on Grok Build is open source
I should try to rob a bank and if I get caught just return the money. No, there needs to be a penalty above what you get, otherwise it encourages people to take the free option of bad behavior. If they get caught they go back as though nothing happened and if they don’t they get a bunch of traces / data.
sarjann··on Codex starts encrypting prompts, uses ciphertext for inference instead
It could also be the case that by the time business adoption picks up a lot they might not be as compute constrained. Depends on rate of growth.
sarjann··on AMD Ryzen AI Halo – $4k AI Dev Kit
No, equities naturally trend up with economic growth. RAM is up because of a supply shock, as new capacity comes on prices will drop, it’s a commodity.
sarjann··on I packed 16 GB of GGUF quants into 1.8 GB, losslessly
Seems pretty useful, often hopping between different quants. I wonder if this would work for different "branches" of a model. E.g. qwen 3.6 35b regular vs abliterated.
sarjann··on Tidal AI Policy
I think this is also a reason why X has gotten worse. They pay people for engagement.
sarjann··on Austria Lobbies EU to Host Anthropic After US Access Curbs
I believe Nvidia chips have a secure way to run your model on other infra.

https://www.nvidia.com/en-us/data-center/solutions/confident...

sarjann··on Google limits Meta's use of its Gemini AI models
Google tends to be very good at vision and smaller/ edge
sarjann··on Founding a company in Germany: €9600, 152 days and I still can't send an invoice
Isn't that what insurance is for? So you don't need to stash away a bunch of idle capital incase something goes wrong.
sarjann··on US holds off blacklisting DeepSeek, more than 100 firms deemed security risks
Text autocomplete can write code, carry out actions (tool calls) and launch cyber attacks. It very much is a matter of national security.
sarjann··on GPT‑NL: a sovereign language model for the Netherlands
The problem is recursive self improvement creating a very difficult gap and the fact that power, compute has a lag from when you invest and when data centers come up.

You also can't just spin up a research team out of nowhere.

sarjann··on GPT‑NL: a sovereign language model for the Netherlands
I wonder with these stories. Why are there so many individual country efforts? We know the scale needed with scaling laws / capital / energy. Most of these countries alone can barely compete (even large groups of them would struggle.

Why don't they work together on it? Companies like Airbus have already been able to do that with aircraft.

sarjann··on Can Europe train a frontier AI model on the compute it owns?
Until they choose to lock their models down (e.g. Fable). Building policy off copying isn't reliable.
sarjann··on I indexed 669 GB of my GoPro videos using my M1 Max computer and local ML models
Asking the important questions
sarjann··on MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second
I don't think token speed matters as much when a lot of tokens are needed to achieve a task. E.g. artificial analysis benchmarks where deepseek v4 is one of the biggest token burners to go through the benchmark.
sarjann··on Google to pay SpaceX $920M a month for compute capacity at xAI data centers
My understanding is google owns a pretty decent stake in spacex (around 6% end of 2025). Booking revenue now will probably help Spacex's valuation.
sarjann··on How LLMs work
I mean there is a little something called compute. And other complexity that comes like writing code to efficiently distribute a model across machines.
sarjann··on Show HN: Uruky (EU-based Kagi alternative) now has Image Search and URL Rewrites
I think it could help to maybe allow 10 searches free (without logging in) just for someone to see what the UI is like or show an example of the results page. It adds a bit of friction for someone to have to make an account and pay just to test the product.

Kagi lets you test it offline (go incognito and try) " Kagi Search is funded by members, not advertisers: built to find what you need, not sell your attention.

Try 50 free searches, and if you love it, sign up for 100 more before choosing a plan. Searches used "

sarjann··on OpenRouter raises $113M Series B
There is also the ability to fallback is one of the clouds degrades in performance.
sarjann··on Replacing a 3 GB SQLite db with a 10 MB FST (finite state transducer) binary
I used something similar for searching over wikipedia pages

https://github.com/sarjann/WikiReader-rs

Was pretty speedy

sarjann··on GPT-5.5 Price Increase: What It Costs
I think it should be tested on goals.

E.g. Crack this puzzle, fix this code so these tests pass. (A human can verify it doesn't cheese things).

sarjann··on UAE to leave OPEC
Isn't the solution in game theory when there are multiple turns to just repeat the move the previous person made.

So if the other side overpumped by x1 amount then you pump an extra x1 the next turn / year (maybe multiplied some reference production factor as they don't all have the same absolute limits).

sarjann··on Stripe valued at $159B, 2025 annual letter
Is there a pure stripe fund though? What if I just want exposure to stripe and not the rest of their portfolio?
← PreviousPage 3 of 3