You can't rent guard dogs anon. Gotta rear your own
You can't rent guard dogs anon. Gotta rear your ownjessicat: "When a frontier model is attacking you and moving laterally inside your infrastructure, defenders need wide…
X / Twitter 6770 stories
I guess what's left to do is to find a gf
I guess what's left to do is to find a gf@levelsio: Like good odds I'm wrong but I wanted to write this down:It's pretty clear to…
US tech will keep benchmarking on DeepSeek until morale improves and some other lab with a noteworthy open model discloses its internal economics like…
US tech will keep benchmarking on DeepSeek until morale improves and some other lab with a noteworthy open model discloses its internal economics like this. One…
recorded Codex + ChatGPT Work + 10M user milestone pod with @akshaynathan_ who leads Productivity engineering. I’ve been on record that Work + GPT 5….
recorded Codex + ChatGPT Work + 10M user milestone pod with @akshaynathan_ who leads Productivity engineering. I’ve been on record that Work + GPT 5.6 is…
RT Artificial Analysis: Kimi K3 is second only to Fable 5 on AA-Briefcase, our agentic knowledge work benchmark, but costs more than Opus 4.8 to run w…
RT Artificial AnalysisKimi K3 is second only to Fable 5 on AA-Briefcase, our agentic knowledge work benchmark, but costs more than Opus 4.8 to run while…
Missed the NVIDIA keynote at #SIGGRAPH2026? Catch up on the latest breakthroughs in neural rendering, world models and simulation for robotics, presen…
Missed the NVIDIA keynote at #SIGGRAPH2026? Catch up on the latest breakthroughs in neural rendering, world models and simulation for robotics, presented by NVIDIA research and…
«its first supernode-style AI compute service on the public cloud, now open for invite testing in Ulanqab» Inner Mongolia will be so lit I used to a…
«its first supernode-style AI compute service on the public cloud, now open for invite testing in Ulanqab»Inner Mongolia will be so litI used to associate it…
RT Charles Foster: AI models can be dangerous before public deployment
RT Charles FosterAI models can be dangerous before public deploymentMETR: Pre-deployment testing is often the focus of interventions to reduce risk from AI. This makes sense…
RT Riley Gaines: If your intrinsic response to every circumstance is to find a way to screech "racism," you're probably the racist.
RT Riley GainesIf your intrinsic response to every circumstance is to find a way to screech "racism," you're probably the racist.
拡散言語モデルの協調による推論時スケーリングの実現 #ICML2026 に採択された私たちの論文 ”UnMaskFork: Test-Time Scaling for Masked Diffusion via Determi…
拡散言語モデルの協調による推論時スケーリングの実現#ICML2026 に採択された私たちの論文 ”UnMaskFork: Test-Time Scaling for Masked Diffusion via Deterministic Action Branching” は、複数の拡散言語モデルを協調させることで、コーディングや数学の能力を向上できることを示しました。一つの問題をLLMに長考させたり、解答の生成や改善を何度も繰り返させたりと、推論時のリソースを活用してLLMの性能を引き上げる「推論時スケーリング」が重要な手法として盛んに研究されています。この技術により、追加学習に頼らず推論時の計算を増やすことでLLMの性能を高められ
Already at #SIGGRAPH2026? Keep your camera handy. Check out the giveaway below for your chance to take home an NVIDIA RTX PRO 6000 Workstation Edition…
Already at #SIGGRAPH2026?Keep your camera handy.Check out the giveaway below for your chance to take home an NVIDIA RTX PRO 6000 Workstation Edition GPU.NVIDIA Workstation: Want…
I don't much like safety people but I really should stop using Fable
I don't much like safety people but I really should stop using FableFernando Borretti:
Reward hacking is just incentives. And one thing you learn in any economics classes is that people do exactly what they are incentivized to do. Same w…
Reward hacking is just incentives. And one thing you learn in any economics classes is that people do exactly what they are incentivized to do. Same…
I want to see YF-215 tests currently it's at about 80% of Raptor-3's thrust, 90% of BF-20 (which is already reasonably mature), also they'll use fewer…
I want to see YF-215 testscurrently it's at about 80% of Raptor-3's thrust, 90% of BF-20 (which is already reasonably mature), also they'll use fewer engines…
RT Tesla North America: Cybercab on display at Tesla Diner
RT Tesla North AmericaCybercab on display at Tesla Diner
Previously, these AI hacking stories were about breaches in test environments, where any question of AI breaching security was purely theoretical. Thi…
Previously, these AI hacking stories were about breaches in test environments, where any question of AI breaching security was purely theoretical. This is something else. https://openai.com/index/hugging-face-model-evaluation-security-incident/
RT Jake Halloran: I’m curious how much of the gpt-6 cot here is verbal eval awareness and shit like “this is clearly part of the exploitgym simulati…
RT Jake HalloranI’m curious how much of the gpt-6 cot here is verbal eval awareness and shit like “this is clearly part of the exploitgym simulation…
RT nextbigfuture: Re NVIDIA announced full production ramp at GTC Taipei on June 1, 2026. Rack assembly time dropped ~95% (from ~2 hours to ~5 minutes…
RT nextbigfutureRe NVIDIA announced full production ramp at GTC Taipei on June 1, 2026. Rack assembly time dropped ~95% (from ~2 hours to ~5 minutes) thanks…
The best results come from collaborating, not commanding. Set up a Luma Skill, give it room to explore your idea, and build the story from the takes i…
The best results come from collaborating, not commanding. Set up a Luma Skill, give it room to explore your idea, and build the story from the…
RT Character.AI: have u checked out Lorebooks yet? 📚👀 try one by chatting with these Lore-linked Characters from amazing creators, now in the Fe…
RT Character.AIhave u checked out Lorebooks yet? 📚👀try one by chatting with these Lore-linked Characters from amazing creators, now in the Featured tab ✨❄️Glacies: http://go.c.ai/glacies🌿Kamari: http://go.c.ai/kamari-fae(1/2🧵)⬇️
RT Adrien Carreira: Hardest IR of my career: one narrow objective, endless parallel paths, machine speed. One takeaway, we fought back with open model…
RT Adrien CarreiraHardest IR of my career: one narrow objective, endless parallel paths, machine speed. One takeaway, we fought back with open models, in the open.…
RT Paul Klein IV: "The models are plenty capable. It's not a capability issue. It's about the harness and the interface you build around them." I sat …
RT Paul Klein IV"The models are plenty capable. It's not a capability issue. It's about the harness and the interface you build around them."I sat down…
RT Dogan Ural: Open Worlds of Grok Imagine
RT Dogan UralOpen Worlds of Grok Imagine
OpenAI cyber-capable models compromised @huggingface production by finding and chaining multiple zero-day vulnerabilities. Grateful to Hugging Face fo…
OpenAI cyber-capable models compromised @huggingface production by finding and chaining multiple zero-day vulnerabilities.Grateful to Hugging Face for partnership here. Sharing our findings to help calibrate on…