Local models + big context = slow. How are you orchestrating "map-reduce" style agent workflows?
I tried running local models (qwen3.6*, ds4 flash, gemma4*, etc) on my mbp pro m5 with 128Gb of unified memory and concluded the bottleneck is context…
Full feed
Feed 28,070 posts
Every story across every category, newest first. Each card links to the original publisher; daily-brief posts open as editorial pages.
RT SpaceX: Our 17th Transporter rideshare mission is targeted to launch tomorrow from California and will deliver 81 payloads to orbit → http://space…
RT SpaceXOur 17th Transporter rideshare mission is targeted to launch tomorrow from California and will deliver 81 payloads to orbit → http://spacex.com/launches/transporter17
b9891
metal: add col2im_1d op (f32/f16/bf16) (#25176) metal: add col2im_1d op (f32/f16/bf16) Gather kernel mirroring the CPU/CUDA path: each output (t_out, oc) reads its ceil(K/s0) source columns…
Blower 5060 Ti’s from Alibaba, good idea?
Asking price is 580 USD on Alibaba, maybe stacking a few of these together for a "power-efficient" system is a good idea? They have 8 pin…
RT Sakana AI: Re Why we built Sakana Translate Blog: https://sakana.ai/translate-release/#English Widely used translation tools often miss the mark ex…
RT Sakana AIRe Why we built Sakana TranslateBlog: https://sakana.ai/translate-release/#EnglishWidely used translation tools often miss the mark exactly where Japanese is hardest. They struggle with the nuance…
AI: The ROI Runway Could Be Long Outside the Tech Sector
Article URL: https://www.apollo.com/wealth/insights-news/insights/daily-spark/ai-the-roi-runway-could-be-long-outside-the-tech-sector Comments URL: https://news.ycombinator.com/item?id=48810533 Points: 3 # Comments: 0
How should I encode both target and feature variable for a multiclass classification? [D]
I am preprocessing a CSV dataset for multiclass classification with XGBoost. My Feature variable contain numerical and categorical values, while the target variable contain many categorical…
Taiganet.com, Home of the WS4000 Simulator
Article URL: https://www.taiganet.com/ Comments URL: https://news.ycombinator.com/item?id=48810501 Points: 4 # Comments: 0
Americans do not belong to the Western civilization.
Americans do not belong to the Western civilization.one dozen rats at a keyboard: Every country on earth would unhesitatingly also bribe FIFA to allow their top…
Teaching LLMs to Recommend and Defer in Underrepresented Epilepsy Care
Specialist epilepsy expertise is scarce in resource-constrained settings, making LLM-based decision support attractive for frontline clinicians managing longitudinal treatment. Such systems must adapt to local prescribing…
they did but compare these and "transformer circuits" side by side and it's pretty clear that Anthropic's research program is more advanced. OpenAI is…
they didbut compare these and "transformer circuits" side by side and it's pretty clear that Anthropic's research program is more advanced. OpenAI is doing engineering, experiments.…
Learning to code is still worthwhile
Article URL: https://stevekrouse.com/learn-to-code Comments URL: https://news.ycombinator.com/item?id=48810439 Points: 59 # Comments: 39
A Review of Anthropic's Global Workspace Paper
The below is a public review Anthropic asked me to write for their new global workspace paper. I recommend at least skimming their paper first.TLDR:I think…
RT McGovern Institute: McGovern neuroscientists have discovered that you don't need language to think logically. People with severe language impairmen…
RT McGovern InstituteMcGovern neuroscientists have discovered that you don't need language to think logically. People with severe language impairments solve complex reasoning puzzles perfectly and brain…
So… anyone copped one of these?
Been almost a year since mass hysteria erupted upon the death of NVIDIAs GPU monopoly. How are your Huawei GPUs? Does CUDA work on them yet?…
RT @sapinker: The data are in: Populism makes countries worse off. "… a country being ruled by a populist is associated with a 10% lower…
RT @sapinker: The data are in: Populism makes countries worse off. "... a country being ruled by a populist is associated with a 10% lower…
b9890
CUDA: remove -sm row, refactor cuBLAS (#24216) CUDA: remove -sm row, refactor cuBLAS fix CDNA + BF16 logic fix bad return fix src0 strides, contiguous requirements…
RT high cagr: Everyone in the AI trade can explain a GPU, but few can explain semiconductor lasers. So I will. Starting with the material. Silicon has…
RT high cagrEveryone in the AI trade can explain a GPU, but few can explain semiconductor lasers. So I will.Starting with the material. Silicon has an…
> Seeds won't require a stargate I really wonder how far DSec + DSpark + everything else have come in the last 10 weeks
> Seeds won't require a stargateI really wonder how far DSec + DSpark + everything else have come in the last 10 weeksAlexander Doria: Virtualized environments…
Reimagining Data Modeling on the Lakehouse: Introducing Vibe Data Modeling
The challenges with Data ModelingIn every analytics stack, the Silver layer is where...
vague sense of gnostic dread and an urge to get blackout drunk after reading locallama
vague sense of gnostic dread and an urge to get blackout drunk after reading locallama
Prefill vs. decoding and local LLM ROI: is prefill underrated?
I'm trying to understand why, when people discuss the ROI of running LLMs locally, they almost always focus on output speed (decoding) and rarely on input…
> ByteDance: installing cackling but yeah! We're still living in the Hopper era. + TPUs/Trainium. "Blackwell class models" like Mythos or GPT-5.x have…
> ByteDance: installingcacklingbut yeah! We're still living in the Hopper era. + TPUs/Trainium. "Blackwell class models" like Mythos or GPT-5.x have virtually nothing to do with…
GLM 5.2 and the coming AI margin collapse
Article URL: https://martinalderson.com/posts/the-upcoming-ai-margin-collapse-part-1-glm-5-2/ Comments URL: https://news.ycombinator.com/item?id=48809877 Points: 31 # Comments: 8