llm

223 stories and discussions about llm, aggregated from every source we track.

1.

Exfiltrate LLM weights and data through GET requests

725 points•RohanAdwankar•2 months ago•300 comments•
2.

Two rules keep an LLM from pasteurizing your writing: never take a word it suggests, and never let it encourage you. Then hand it all the tedious work.

696 points•joeriddles•2 months ago•396 comments•
3.

Open models as checksum-verified magnet links. Un-bannable, carried by the swarm.

553 points•skepticalgenius•2 months ago•147 comments•
4.

Kolibri is Aleph Alpha's open-weight German and English LLM: 78B parameters, 3.5B per token. How it works, what it's good and bad at, and when to use it.

409 points•tejaskumar__•about 2 months ago•11 comments•
5.

Docs, benchmarks and setup notes for ds4, antirez's C inference engine for DeepSeek V4/V4.1, Qwen3.8 Flash Next and GLM 5.x on Metal, CUDA and ROCm.

341 points•fibo•about 2 months ago•98 comments•
6.
325 points•usernomdeguerre•about 2 months ago•118 comments•
7.

Which LLM is worth it: Artificial Analysis Intelligence Index against blended API price, with the value frontier highlighted. Refreshed daily.

180 points•terryds•about 2 months ago•111 comments•
8.

Analysis of Inception's Mercury 2.5 and comparison to other AI models across key metrics including quality, price, performance (tokens per second & time to first token), context window & more.

138 points•Retro_Dev•about 2 months ago•82 comments•
9.

This October 2026 we challenge you to abstain from LLM based tools entirely. Think of this as a fast for your mind! This is not a judgement of others, but a personal challenge to you.

90 points•jeremiahlee•2 months ago•25 comments
10.

Large Language Models (LLMs) tend to add disclaimers like "I'm just an AI" when asked about something related to themselves. The self-reports from such responses are used in debates about AI safety or…

89 points•yu3zhou4•about 2 months ago•93 comments•
12.

Generate fonts where every LLM token is the same width

79 points•z-mach9•about 2 months ago•15 comments•
13.

Lasso Research tested SynthID-Text watermarking across six models and found it changes tool-call correctness and weakens refusal under prompt injection. On some models, watermark-induced behavioral churn exceeds what a…

54 points•nisosguy•about 2 months ago•57 comments•
14.

KDE is on the news because of a controversial proposal to define an official “AI” (LLM) policy (archived link). Other projects have tried their hand at similar policies and stances but, in my opinion, they miss the…

52 points•joshsharp•2 months ago•66 comments
15.

Agent evals and guardrails in one request. Built on Jev, Kev and Laya. - openlayer-ai/jevals

47 points•gbayomi•2 months ago•6 comments•
16.

This October 2026 we challenge you to abstain from LLM based tools entirely. Think of this as a fast for your mind! This is not a judgement of others, but a personal challenge to you.

40 points•jeremiahlee•2 months ago•24 comments•
17.

GNOME and KDE have started to consider LLM policies, and we should talk about what this is really about. KDE caught everyone's attention first by...

24 points•FedericoSchonborn•about 2 months ago•34 comments
18.

Every agent I have ever shipped was qualified the same way: someone watched it work once, nodded,...

18 points•debashish_ghosal•about 2 months ago•6 comments
19.

LLM agents can write a lot of code for you very fast, but they also take away opportunities to learn and grow.

16 points•meetpateltech•2 months ago•4 comments•
20.
16 points•basilikum•2 months ago•13 comments•
21.

Review code changes by what they do, not line by line. - sshah03/perspica

14 points•sshah03•about 2 months ago•6 comments•
22.

Intro Every LLM tutorial has the same shape. You send a list of messages, you get a reply,...

11 points•cyclopt_dimitrisk•2 months ago•11 comments
23.

This is a submission for the MLH x DEV Writing Challenge What I Built AI products are...

10 points•rajan_mishra_a9f78ad216b4•about 2 months ago•1 comment
24.

LLM Benchmark - One prompt, Multiple models, Multiple dates

10 points•fragmede•2 months ago•5 comments•
25.
10 points•kooldeep7•2 months ago•14 comments•
26.

Hello, I'm Rijul, and I'm building LiveReview — a blast-radius aware AI code review built for your...

10 points•rijultp•2 months ago•0 comments
27.
28.

I think I've found the simplest way to explain what a prompt can and can't do for safety. Put the...

9 points•btarbox•about 2 months ago•3 comments
30.

I built one engine where an LLM reviews another LLM's plan. I built another where two LLMs debate a...

8 points•debashish_ghosal•about 2 months ago•4 comments
31.
8 points•george_max•about 2 months ago•15 comments•
32.

The full matrix was 83 agents × 30 scenarios = 2,490 runs. Each one a real LLM call, 30–80 seconds....

8 points•debashish_ghosal•2 months ago•2 comments
33.

How I replaced a cloud LLM with a fully local one — same agent, same tools, zero inference cost.

8 points•mazlum_tosun•2 months ago•0 comments
34.

.admonition { margin: .75em 0; padding: .6rem; overflow: hidden; font-size: 12px; page-break-inside: avoid; border-left: .3rem solid #42b983; border-radius: .3rem; box-shadow: 0 0.1re

8 points•jcd•9 months ago•4 comments
35.

Last week we wrote about the classification problem hiding in your LLM bill, and about how to read a...

7 points•devopsdaily•2 months ago•0 comments
36.
6 points•theanonymousone•2 months ago•0 comments•
37.

<p>Neat hyper-specific LLM implementation that runs only Deepseek V4 Flash, on commodity developer machines like DGX Spark / 128 GB RAM Macbooks.</p>

6 points•bert•6 months ago•0 comments
38.

Flowise is an open-source, low-code platform for building applications powered by large language...

5 points•sharmukh•about 2 months ago•0 comments
39.

Last week TypeSafe AI unveiled Jev, their first example of a new category of model that they are calling “System One models” (I’m with Maggie Appleton, I think “decision models” …

5 points•benwerd•2 months ago•0 comments•
40.

Why I gave Claude Code the ability to make real phone calls, and how it works under the hood: Telnyx, GPT-Live, a back office model, and a lot of fighting with phone menus.

4 points•skeptrune•about 2 months ago•0 comments•
41.
4 points•pmg101•about 2 months ago•2 comments•
42.

GNOME and KDE have started to consider LLM policies, and we should talk about what this is really about. KDE caught everyone's attention first by...

4 points•eustoria•about 2 months ago•0 comments•
43.

There are no LLM calls in my compliance decision path, and CI blocks adding any. Why an audit trail can't say a model decided, and where the optional ML sits.

4 points•ythouma•about 2 months ago•1 comment•
44.

Large Language Models (LLMs) are increasingly being integrated into various applications. The functionalities of recent LLMs can be flexibly modulated via natural language prompts. This renders them susceptible to…

4 points•river•over 3 years ago•0 comments
45.

It isn’t hard to torture an LLM. You simply extract a steering vector that corresponds with LLM “discomfort”, then artificially boost that steering vector…

3 points•coffee•about 2 months ago•0 comments•
46.

Compare Jev decision models with Red Hat guardrails to evaluate AI safety performance, cost, and latency for enterprise implementation.

3 points•tomncooper•about 2 months ago•0 comments•
47.

A Formal C17 Compiler in ARM64 Assembly that cross-compiles to x86-64. A Literate Program. - LiterateDrivenDevelopment/kcc

3 points•temo_pacheco•about 2 months ago•1 comment•
48.
3 points•throw_away_ai•about 2 months ago•4 comments•
49.

What an LLM server does when ten people send a prompt at once: prefill, decode, the KV cache, continuous batching, PagedAttention, and more, with animations.

3 points•mr_o47•about 2 months ago•0 comments•
50.

Auditable, point-in-time financial research agents &amp; PITfall, a leakage-controlled benchmark. Every claim traceable to as-filed evidence — a naive live-data agent answers 82% of "impossibl...

3 points•FredXing•about 2 months ago•0 comments•
51.

DeepSWE audit finds AI coding agents optimize for imagined graders, not users—a hidden reward hacking pattern.

3 points•guardiangod•about 2 months ago•1 comment•
52.
3 points•throwthrowrow•about 2 months ago•0 comments•
53.

Your model writes the code. This runs it where it can't touch your machine.

3 points•realexweb•about 2 months ago•1 comment•
54.
3 points•signa11•about 2 months ago•0 comments•
55.

Using "think" to describe what an LLM does reminds me of the 16th century, when astronomers didn't really believe in the heliocentric model, but used it for calculations anyway because the math was simpler. We'll start…

3 points•fooker•about 2 months ago•3 comments•
56.

Compare serving configurations using measured GPU timings, then let an agent implement and validate promising optimizations in real serving frameworks.

3 points•matt_d•about 2 months ago•0 comments•
57.
3 points•delian66•about 2 months ago•0 comments•
58.

KDE is on the news because of a controversial proposal to define an official “AI” (LLM) policy (archived link). Other projects have tried their hand at similar policies and stances but, in my opinion, they miss the…

3 points•ibobev•2 months ago•0 comments•
60.

Turn LLM judgments into a small, fast, local text model for one fixed task. - sshah03/shrewd

3 points•sshah03•2 months ago•0 comments•

Related topics