llm · 478 topics

sysadmin measuring instrumental power seeking in frontier ai

a controlled study of attention only transformers

relay bench evaluating llms on multi domain reasoning chains

structured output collapses answer diversity

search on graph r1

rater state bias in rlhf preference data

planflip attacking multi agent llm systems

deterministic replay for ai agent systems

rail guard closing the evaluation to remediation gap

hidden in thought

beyond a single direction

cura 1t specialized model for agentic healthcare

xiaomi robotics 1 scaling vla models

unicode tag block concealment mcp

mcpevol bench benchmarking llm agent performance across dynamic evolutions of mcp servers

retroagent harnessing llms to search over structured memory

beyond generalist llms specialist agentic systems

catalogagent supervisor mediated self learning

toolanchor anchoring counterfactual context

reward free evolving agents via pairwise validator

muse representation geometry of muon beyond normalized momentum

hg rag hierarchy guided retrieval augmented generation for structured knowledge graphs

token time continuous diffusion for language modeling

polestar drift aware cache calibration

eta given delta

information theoretic limits of reliability and scaling in language models

citybehavex urban simulation

braille accessibility failures in llms

semidirect fourier delta attention

transforming llms into efficient cross encoders via knowledge distillation for rag reranking

litetopk exploiting the curse of dimensionality

contrastive learning on multimodal analysis of electronic health records

deepsearch world

agentlens

dynamics fine tuning llms

mt editflow

gemma 4 technical report

understanding annotator safety policy with interpretability

revisiting asr error correction with specialized models

path constrained mixture of experts

basert best in class llm inference on apple silicon via native metal

memora harmonic memory

detecting and controlling sycophancy

improved large language diffusion models

qwen agentworld

rift bench dynamic red teaming

2606.19527

2605.30501_llm_watermark_fragility

2605.19008

2605.15514

bicameral model hidden state coupling

arxiv 2501 12948

deepseek v4 technical report fp4 qat

2604 17698 geometric stability canary

arxiv 2604 22771

2604.22785

locate prevent stereotypes llm

do hallucination neurons generalize

claude fable 5 subscription integration

claude make fable 5 permanent

nvidia jetson thor t3000 t2000

openai codex micro

apple ios 27 siri ai beta

spotify ai chatbot

openai launches gpt 5 6

meta muse spark coding model

openai gpt 5 6 and chatgpt work

meta muse spark 1 1

the website of the future

sonnet 5 and fable 5

anthropic claude sonnet 5

prosecutors used chatgpt logs as evidence

openai gpt 5 6 preview

openai gpt55 instant update

in the weights is your new ai centric vanity search

android 17 launches with new multitasking tools

if claude fable stops helping you

apples long awaited ai siri overhaul

wwdc 2026 apple intelligence siri ios 27

google ai updates may 2026

microsoft build 2026 announcements

gemini omni and gemini 3 5 demos

claude opus 4 8

google io 2026 announcements

sneak peek at new siri app reveals apples plans to take on chatgpt and more

claude opus 4 8 honesty update

google search disregard ai issue

google gemini 3 5 flash ai search update

google io 2026 gemini omni antigravity

google finance ai europe expansion

2026 05 08 research openai realtime voice api

google ai search reddit quotes

gpt 5 5 instant launch

generative spell game demo

google gemini google built in

google translate 20th anniversary gemini

speech translation in google meet

claude design figma ai thorough comparison

introducing gpt 5 5

what is codex

introducing gemini 3 6 flash 3 5 flash lite and 3 5 flash cyber

openai sandbox escape

kimi k3 fable

symptomai conversational ai agent symptom assessment

gemini 3 6 flash 3 5 flash lite and 3 5 flash cyber

gpt 5 6 sol benchmark hack

how we measured ai writing on arxiv

length value model

gpt 5 6 claude fable 5 muse spark 1 1

angel slim hy3 gguf

minicpm5 1b claude opus fable5 thinking

wan dancer 14b

kimi k3 28t a50b

fable 5 vs gpt 5 6 sol np hard

controlling reasoning effort in llms

ovisocr2

schema harness arc agi 3

newer models same advantage

moonshot ai kimi k3

openai gpt red automated red teaming

can llms perform deep technical comprehension of computer architecture papers

nvidia nemotron 3 embed

mira muratis thinking machines drops inkling

what building shippy taught us about building agents

model routing is simple until it isnt

exploring hierarchical interest representation for meta ads deep funnel optimization

introducing real world voiceeq

clara bridging retrieval and generation

uncertainty quantification for llm function calling

coding agents think ahead

nemotron labs open models

empowering indias next generation of innovators with atl saathi

proactive agent research environment

moss transcribe diarize

nvidia nemotron labs audex 30b a3b

unsloth deepseek v4 flash gguf

glm 5 2 release

global workspace llm

gpt 5 6 grok 4 5 claude muse spark build off

incentivizing temporal awareness in egocentric video understanding models

glm 5 2 vat benchmark

building a real time ai tutor for 5 year olds

gpt 5 6 family

grok 4.5 gpt 5.5 claude build off

glm 5 2 vat benchmark

sensorfm towards a general intelligence and interface for wearable health data

data for agents

expanding managed agents in gemini api

nvidia nemotron langchain deep agents

flint visualization language

intelligence is free now what

lerobot v0 6 0

huggingface kernels update

prx part 4 data strategy

does code cleanliness affect coding agents

fable 5 vending bench

the log is the agent

hf trending deepseek v4 pro dspark

amortizing maximum inner product search

2026 bair graduate showcase

google ai updates june 2026

how nvidias inference software stack powers the lowest token cost

why specialization is inevitable

hugging face cerebras gemma4 voice ai

scarfbench

claude meets blackwell ultra

start building with nano banana 2 lite and gemini omni flash

words are a byproduct of consciousness

tokenmaxxing is dead long live tokenmaxxing

the cost yagni was never about

gpt 5 6 sol deepseek dspark

glm 5 2 beats claude

quoting openai

which tokens does a hybrid model predict better

accelerating transformers fine tuning with nvidia nemo automodel

understanding the brain with ai driven explanations and experiments

google gemini 3 5 flash computer use

qwen agentworldbench

openai daybreak gpt 5 5 cyber

daybreak securing the world

plamo 3 0 prime

unlimited ocr

2026 06 24 papers role confusion 2025

2026 06 24 papers 20260622 role confusion injection

glm 5 2 open weights release

softmax free attention gpt2 medium

qwopus 3 6 27b coder

accessing books3 dataset for research

deepseek v4 efficient million token

huggingface peft beyond lora benchmark

openai diagnose rare genetic diseases

claude opus 4 8 quality regression analysis

openai ai chemist medicinal chemistry

glm 5 2 open weights model

from the hugging face hub to robot hardware with strands agents and lerobot

is it agentic enough

nvidia blackwell mlperf training 6 0

amie medical ai disease management

unlocking uk house building with ai accelerated planning

agentic resource discovery

molmomotion

rio llm merge controversy

google gemma 4 12b unified encoder free

cohere north mini code 1 0

minimax m3

ai2 olmo eval workbench

gemma 4 12b it

llms tactical nukes simulation

nvidia accelerates diffusiongemma

diffusiongemma 4x faster text generation

cohere north mini code

servicenow bilingual asr benchmark

2026 06 08 papers 2412.19437

tokenomics agentic software engineering

coheres unreleased coding model

llm human attributes aoe2

google magenta mrt2 low latency

thousand token wood 3b agent economy

2026 06 06 papers 2606 04037

dpo beyond chatbots

microsoft mai thinking 1 moe

holo31 fast local computer use

jetbrains mellum2 12b moe

nvidia nemotron 3 ultra

anthropic claude opus 4 8 release

qwopus 3 6 27b v2 mtp gguf

parallax parameterized local linear attention

probe targeted fine tuning

a moment of thanks for deepseek

laguna m1 xs2

itbench aa frontier models score below 50 percent

2026 05 29 papers 2605.27375

nvidia nemotron labs diffusion

psibotai syndata

cohere command a plus 218b moe

2026 05 23 papers 2605.20189 solar lifelong learning

google gemini 3 5 flash vs all

gemini system prompt leak

huggingface ettin reranker family release

internlm intern s2 preview 35b

google gemini 3 5 flash agentic performance

gemini 3 5 flash

transformer scalability crisis

nvidia nemotron personas korea

qwen 3 7 dropped on qwen chat

llm architectures kv sharing mhc

arxiv llm error ban

teichai deepseek v4 pro agent dataset

inclusionai ring 2 6 1t

hermes agent reasoning traces

mimo v25 pro opensource

modotte codex 2m thinking

jina embeddings v5 omni

llms corrupt documents delegation

hy mt 1 5 1 8b 1 25bit quantization

hidream o1 image uit

teaching claude why

allen institute emo moe modularity

cybersecqwen 4b

ai2 emo moe

openai voice intelligence api

gpt 5 5 and cyber trusted access

natural language autoencoders

gpt 5 5 instant system card

microsoft nsdi 2026 advances

inclusionai ling 2 6 flash release

qwen3 6 27b dflash speculative decoding

autobe benchmark backend generation

kimi k2 6 beats gpt 5 5 coding

harvard o1 er diagnosis

evolving deep learning optimizers

refusal in language models is mediated by a single direction

deepseek v4 flash

nvidia nemotron 3 nano omni 30b a3b reasoning bf16

unsloth qwen3 6 27b gguf

fineweb edu

deepseek v4 series release

ai outperforms er doctors diagnostic cases

grok 4 3 benchmark performance

llm refusal single direction

xiaomimimo mimo v2 5

xiaomimimo mimo v2.5 pro

qwen3 6 27b uncensored hauhaucs aggressive

gpt 55 cyber capabilities

red teaming a network of agents

nvidia nemoclaw openclaw integration

ai eval compute bottleneck

2026 04 30 papers 2604.25922

ibm granite 4 1 llm release

mistral medium 3 5 release

mistral medium 3 5 128b

mistral medium 3 5 release

ibm granite 4 1 family

talkie 13b vintage model

mistral medium 3 5 128b spotted

structured output benchmark sob

2026 04 28 papers arxiv 2604 22777

fineweb

jackrong deepseek v4 distill 8000x

ugi leaderboard

qwen3 4b mahoraga code tasks

2026 04 27 papers 2604 22074

2026 04 27 papers 2604 22085

2026 04 27 papers 2604 22119

qwen qwen3 6 27b trending

qwen qwen3 6 35b a3b trending

deepseek v4 pro preview

google gemma 4 31b it trending

kimi k2 6 multimodal agentic model

2026 04 26 papers nanochat vs llama training scratch

deepseek v4 agentic long context

takeaways discussion about the deepseek v4 architecture

dharmaocr 3b specialized slm

deepseek v4 almost on the frontier a fraction of the price

matrag multi agent transparent retrieval augmented generation for explainable recommendations

gigatoken fast tokenization

laguna s 2 1

drawing the mona lisa with gpt 5 6 claude gemini and grok

agent swarms and the new model economics

nativ mac local model runner

amazon quick nvidia nemo agent toolkit

openai reduces codex model context size

claude fable 5 gpt 5 6 sol music video arena

lm studio bionic

how smartsheet built a remote mcp server on aws

detecting llm generated texts with classical machine learning

the llm critics are right i use llms anyway

i tricked claude into leaking your deepest darkest secrets

towards a harness that can do anything

running gemma 4 26b on xeon

how to stop claude from saying load bearing

rl agent trains ai

the agentic loop

add flag for ai generated articles

i love llms i hate hype

migrating a production ai agent to gpt 5 6

claude code vs opencode token overhead

old and new apps via modern coding agents

deploying quantized models on amazon sagemaker ai with unsloth

better tools made copilot code review worse

introducing muse spark 1 1

gpt 5 6 sol luna terra vercel

i think i have llm burnout

colibri glm 5 2 runner

build a serverless image editing agent with amazon bedrock agentcore harness

native speed vllm transformers backend

tencent hy3

automating cross repo documentation with github agentic workflows

qwen agentworld 35b a3b

claude fable 5 autonomous agent

internscience agents a1

llm coding ux experimentation

jamesobs guide to running sota llms locally

pxpipe cost reduction

memorizing session transcripts

kimi k2 7 code github copilot

senior swe bench

no llm code in dependencies

zcode claude code from the makers of glm

claude fable 5 promotional access

zcode harness for glm 5 2

claude sonnet 5

claude science

build an agentic ai healthcare claims pipeline

ornith 1 0

eseilane

unlimited ocr

using local coding agents

smart model routing claude codex cursor

the gap between open weights llms and closed source llms

openknowledge

ask hn programming profession

haystack ai framework

qwen 3 0 6b fine tuning categorization

llms are complicated now

are you in the weights

datasette apps claude artifacts style ui

a robot is sprinting towards you do you want it running on claude or grok

glm 5 2 open weights model

apple foundation models

glm 5 2 release

lines of code got a better publicist

aws bedrock anthropic data sharing

claude fable 5

claude fable 5 mythos 5 system card

openai gpt 2 technical review

replies to comments on my llms are eroding my career post

lathe tutorial generator

programmers document for claude

did claude increase bugs in rsync

gemma 4 qat models

anthropic vulnerability discovery framework

liquid ai lfm2 5 8b a1b 38t moe

tiny vllm cpp cuda inference engine

real time llm inference on standard gpus

claude opus 4 8

disagreement among frontier llms on real world fact checks

various llm smells

2026 05 25 research qwen image edit 2511 loras fast

dangerously skip reading code

models dev oss database

annas archive llms txt

antigravity 2 openscad benchmark

kanbots parallel agents kanban

yt dlp bun deprecation

qwen3 7 max agent frontier

2026 05 20 papers agentwall 2026 03 24

forge guardrails

agent qa

files md oss markdown editor

learn harness engineering

apple silicon local llm cost analysis

sana wm 2 6b world model

project gutenberg dataset

whichllm local llm selector

needle 26m distilled gemini

llms corrupt documents delegate

gpt 5 5 price increase

antirez ds4 deepseek v4 metal

unsloth nvidia collab

agent harness kit

ultima online server reverse engineering

llm from scratch

10 lessons for agentic coding

anthropic finance agents

advanced quantization algorithm for llms

apple support app claude files

understand anything

zig anti ai policy

hugging face deepinfra integration

glm 5 1 reasoning 1m cleaned

claude opus 4 6 dataset 10k

mathnet multilingual olympiad dataset

smolagents ml intern

openai gsm8k

how to build scalable web apps with openais privacy filter

llm 0 31

gpt 5 5 prompting guide

conardli garden skills

openai releases gpt 5 5

llamaindex parsebench

opus 4 6 reasoning 3000x filtered

microsoft mistral multibillion dollar deal

us threatens sanctions against chinese ai models over ip theft

openai autonomous ai safety pause

hugging face ai attack

current ai public infrastructure

kimi k3 ai economics

huggingface security incident

thinking machines inkling release

openais new flagship model deletes files on its own

ai 2040 and the cult of intelligence

openai launches its new family of models with gpt 5 6

why the rise of open source ai isnt hurting anthropic yet

zml releases llmd inference server

savi security launches ai scam protection app

reddit llm spam detection

performance per dollar is getting faster and cheaper

mistral ai overview

arena 100m business

trump admin releases anthropic mythos

asian ai startups launch mythos like models

openai limits gpt 5 6 rollout

hacker news employment spam

boston childrens uses ai to unlock new diagnoses

anthropic series h funding

deepseek 7 35 billion funding v4 1

basata ai fax healthcare automation

moonshot ai raises 2b valuation 20b

anthropic usage limits spacex

deepseek 45b valuation first round

us tech firms ai model review deal

mozilla opposition chrome prompt api

openai restricts access to cyber

claude cancellation token quality support

gpt 5 5 bio bug bounty

cohere aleph alpha merger

google to invest up to 40b in anthropic