Genert Org

That polygot software guy before LLMs took over.

[Genert Org @_genert]

@Aaronontheweb You have fallen prey to Claude marketing.

[Genert Org @_genert]

Just wired up Qwen3.6-27B with MTP on llama.cpp. One flag (--spec-type draft-mtp), ~1.5-2x throughput, identical outputs. Local inference keeps eating the moat.

[Genert Org @_genert]

@mitchellh LoC and deploys per day aren’t useless metrics, not for model providers because they push you to consume more. Maybe there will be a day where we aim to create most token efficient code and delay change as far as possible.

[Genert Org @_genert]

@viljararakas Doesn't stop the ever steeping retirement curve and eventual collapse of pension system. Pretty sure this will be snugged under the rug in the next election cycle.

[Genert Org @_genert]

@dhh I never understood the license issue on the code generation. At the end of the day it's token generated code converted to the most efficient machine code via compiler and fed as instruction set to CPU. What is left there to copyright?

[Genert Org @_genert]

@elonmusk Any plans to open source it?

[Genert Org @_genert]

@Talvitp @viljararakas 2,5 miljardit? Alusta endast... Enne kui Malthuse maailmavaadet jagad, tasub meeles pidada, et kogu maailma rahvastik mahub New Yorgi Central Parki ära, kui kokku pakkida nagu kilud karbis. Meid on tegelikult vähe ja see saab suureks probleemiks.

[Genert Org @_genert]

@nateberkopec It is not single variable optimization. This can solve an issues I have had with online learning.

[Genert Org @_genert]

@karpathy All what was left to do was to delegate feedback to agents. Getting this to work on the scale of massive models is going to be operational challenge but I think in a year we will get there.

[Genert Org @_genert]

@Rafael_L_Spring lmao

[Deedy @deedydas]

So many startups think their engineers are "cracked" but have no idea what that really means.

This team of 5 19yr olds built a 30 petabyte storage cluster in SF for ~$500k to get a 40x cheaper AWS S3 as a side quest to store 90M hours of video.

Now, that's cracked.

[muke @muke1010101]

Accidentally stumbled across some insane Gem5 stuff. Turns out the Chinese have just been pushing the SotA in hardware simulation without telling anyone.

[Genert Org @_genert]

@matteopelleg I am sure that within 3 years, any new iPhone will run local LLM that is twice "smarter" than Opus 4.6. New architecture, separate chips for LLM, and battery improvements will definitely allow it.

[Genert Org @_genert]

@taurialas BHF võlakoormust, võlakirju ning portfooliot vaadates tekib arusaam, et BHF strateegia on tänaseks olnud "teeme kõik, mida EfTEN teeb aga vastupidi!".

[Genert Org @_genert]

@jaanus Censorship free fine-tuned models exist to mitigate these risks. As long these models are open source, we can do it - https://t.co/PFhDG1aLOx

[John Carmack @ID_AA_Carmack]

256 Tb/s data rates over 200 km distance have been demonstrated on single mode fiber optic, which works out to 32 GB of data in flight, “stored” in the fiber, with 32 TB/s bandwidth. Neural network inference and training can have deterministic weight reference patterns, so it is amusing to consider a system with no DRAM, and weights continuously streamed into an L2 cache by a recycling fiber loop. The modern equivalent of the ancient mercury echo tube memories. You would need to pipeline a bunch of them to implement modern trillion parameter models, but fiber transmission may have a better growth trajectory than DRAM does today, so it might someday become viable.

Much more practically, you should be able to gang cheap flash memory together to provide almost any read bandwidth you require, as long as it is done a page at a time and pipelined well ahead. That should be viable for inference serving today if flash and accelerator vendors could agree on a high speed interface.

[Genert Org @_genert]

@Brad_Setser @ExanteData ECB just has to add a 10% risk weight to US Treasuries in Solvency II. EU insurers would be forced to dump US debt instantly. Trump is only influencable through treasury markets.

[Genert Org @_genert]

@NatRothschild1 @Simon_Nixon Entering the defense market is much easier in the US than in Europe. Europe struggles with market fragmentation, political interference, and resistance to innovation. Consequently, Europe is lagging behind the rest of the world in critical defense capabilities.

[Genert Org @_genert]

@NatRothschild1 @Simon_Nixon Europe values its security, but the defense sector suffers from underfunding and high entry barriers compared to the US. Procurement is often driven by politics rather than the free market. Policymakers must urgently address these issues.

[Bidipta Sarkar @bidiptas13]

Evolve at the hyperscale!
Work co-led with Mattie Fellows and Juan Agustin Duque.
Made possible by #Isambard and AIRR

🌐 Website: https://t.co/JLluUPmuWc
📝 Paper: https://t.co/JonCoelVIT
💻 Code: https://t.co/moyzZfEg9s
🥚NanoEgg : https://t.co/LLmgnnm6eU (train in int 😉)