search
large language models
Trends
- 1Stanford paper claims method could reduce reliance on Nvidia memory chips●Is This the End of LLMs? This Stanford Paper Could Make Nvidia's Memory Chips Unnecessary
A Stanford research paper is drawing attention with the claim that a new technique could make Nvidia's high-bandwidth memory chips unnecessary for running large language models. The paper has sparked debate in the AI and semiconductor communities over whether such a shift is realistic, with some seeing it as potentially disruptive to Nvidia's business and others urging caution about unproven research claims.
- 2Cheaper model with Opus escalation cuts AI costs fourfold▼A Jev that can ask Opus for help gives Opus accuracy at 4x lower cost
A new write-up from Hightouch describes an AI setup where a smaller, cheaper model handles routine work and escalates harder cases to Anthropic's Opus. According to the company, this routing approach matches Opus's accuracy while costing four times less, using LLM judges to decide when a task needs the stronger model.
- 3New tool runs 284B-parameter DeepSeek model without GPU●Llama-modchip: 284B DeepSeek on 64GB RAM, no GPU
A project called Llama-modchip claims it can run DeepSeek's 284-billion-parameter model on a machine with just 64GB of RAM and no graphics card, removing the expensive GPU hardware normally required for large language models. Developers on Hacker News are discussing how the technique works and what it means for running powerful AI locally on ordinary computers.
- 4German user eyes Simplebanking for AI-linked account access●RE: https:// troet.cafe/@nSonic/11758658713 6488076 Ich werde mir wohl mal Simplebanking anschauen. https:// simplebanki
A German-language social media user says they plan to try Simplebanking, a German service that aggregates multiple bank accounts free of charge. It offers an interface, including MCP support, allowing users to connect their own large language model and work with their financial data directly. The post highlights growing interest in combining open banking aggregation tools with personal AI assistants.
- 5Man wins first religious exemption from mandatory AI use at work●Interesting! "Man granted first-of-its-kind # ReligiousExemption against mandatory # AI use at work- # Humanist argues t
A US employee has reportedly been granted a first-of-its-kind religious exemption after refusing to use mandatory AI tools at work, arguing that offloading his thinking to large language models violates his secular humanist beliefs. The accommodation is said to rest on Title VII, the federal civil rights law protecting religious practice, and commenters are debating whether non-theistic beliefs qualify.
- 6Uncracked City detective game casts lying LLM suspects●Show HN: Uncracked City, detective game where you question LLM suspects that lie
A new browser game called Uncracked City puts players in detective role, interrogating suspects powered by large language models that deliberately lie during questioning. Players must spot contradictions and crack cases by pushing characters for the truth. The project launched as a Show HN submission, drawing over a hundred upvotes and early attention from developers and AI enthusiasts discussing how convincingly the models deceive players.
- 7Lawyer reflects on AI's leap from early neural nets to LLMs●When I went to # law school, the state-of-the-art # AI was still Rumelhart’s BP and Kohonen’s SOM. That was a roundabout
A lawyer and academic notes that when he attended law school, the cutting edge of artificial intelligence consisted of Rumelhart's backpropagation and Kohonen's self-organizing maps. He argues law students of his era learned to read and write the way lawyers do rather than relying on tools, and observes that today most lawyers he knows use large language models in their daily work.
- 8Lawsuit Against Google Over Alleged AI-Assisted Suicide Planning Draws Attention●Thanks to @ oldbookillustrations I found a great metaphorical illustration for a piece on schools, LLMs and children. Th
A writer preparing a piece on schools, large language models and children is using a historical metaphorical illustration found via a public-domain archive, prompted by a newly reported case against Google. The lawsuit involves allegations that AI assisted in suicide planning, adding to ongoing scrutiny of how chatbots handle young or vulnerable users.
- 9GPUWorks launches as lightweight GPU cluster manager without Kubernetes●Show HN: GPUWorks – A 40 MB GPU cluster manager for vLLM/SGLang, no K8s needed
A new open-source tool called GPUWorks is being introduced as a 40 MB GPU cluster manager for running vLLM and SGLang inference workloads without requiring Kubernetes. The project, available on GitHub from flitai-io, is positioned as a simpler, lighter alternative for teams managing GPU clusters for large language model serving.
- 10
Revolut is reportedly shifting to a self-hosted deployment of GLM, the large language model developed by Z.ai (formerly Zhipu AI). The fintech company, one of Europe's largest banks, appears to be bringing AI inference in-house rather than relying on external providers. The move is drawing attention among developers discussing cost, control and data-privacy trade-offs of self-hosting frontier Chinese open models.
- 11Interactive notes site explains RLHF, attention and LoRA●Show HN: Interactive notes on RLHF, attention, LoRA and LLM evals
A developer has launched a website offering interactive notes explaining core machine learning concepts behind large language models, including RLHF, attention mechanisms, LoRA fine-tuning and LLM evaluations. The material uses visual, hands-on explanations rather than plain text, and it is drawing attention from readers interested in accessible learning resources for AI fundamentals.
- 12Developers debate hardware for running local LLMs●Ask HN: What hardware are you using to run local LLMs?
A discussion thread on Hacker News is asking people what hardware they use to run large language models locally. Respondents are comparing consumer GPUs, Apple Silicon Macs, and dedicated workstations, weighing memory requirements, inference speed and cost against cloud API pricing. The conversation reflects growing interest in running capable AI models on personal machines rather than paying for hosted services.
- 13Can software startups survive the era of large language models?●In LLM era, can software startup survive?
A discussion is underway about whether traditional software startups can survive as large language models make it cheaper and faster to build products. The core question: when a few prompts can reproduce a startup's core feature, what remains defensible? Participants are weighing whether LLMs are an existential threat to small software companies or simply a new tool that shifts where value and moats lie.
Repos
- synchopate/llama-modchip Run 284B DeepSeek-V4-Flash (<59GiB RAM: 41.1 cold pp, 9.1 N=1 tg, up to 25.3 spec tok/s) & 30B MoEs (22.4 N=1 / 1
- hyperionaidev/smolmaxxing maxxing the ability to run different LLMs on the smallest devices possible
- JustVugg/colibri A hierarchical inference engine for frontier MoE models: hot experts on the GPU (CUDA, Vulkan), warm ones in RAM, cold o
- ggml-org/llama.cpp LLM inference in C/C++
- shiyu-coder/Kronos Kronos: A Foundation Model for the Language of Financial Markets
- spotify/portal-ai-plugins
- emmilco/us-tax-model US federal income tax forms and instructions as cited, executable, machine-checkable data.