Your curated collection of saved posts and media
Drop 2/15 Are we actually measuring the things we ultimately want our voice AI models to achieve? The best AI voices are becoming hard to tell apart. From each other. Not from humans. ๐ Introducing Turing TTS: an evaluation that measures how often human listeners mistake AI-generated speech for real human speech. We tested leading TTS models against real human speech. Even the best model fooled listeners only ~10% of the time. With Turing TTS, we hope to contribute to a broader discussion within the research community about how we measure progress in voice AI. If one of our goals is to develop speech that is indistinguishable from human speech, how well do our current evaluation methods capture progress toward that goal? We donโt claim that Turing TTS is the right way forward, or that human indistinguishability should be the goal of every TTS system. Rather, we hope these findings encourage the community to collectively examine whether blind pairwise human evaluations of naturalness are sufficient, and what else we might need to measure. Results are now live on @voicearena_ai.
Building safer AI agents takes more than model safeguards. @perplexity_ai open sourced SPACE, the agent sandbox platform behind Computer, giving developers more control over where agents run code and what they can access. Cheers to more teams open-sourcing tools to build safer autonomous AI agents.
We're open sourcing SPACE, the sandbox platform that runs every task in Perplexity Computer. SPACE gives each agent task its own Firecracker microVM and controls credentials and outbound traffic from outside the VM. Available now under Apache 2.0: https://t.co/DzSmq6o1xA
World models like Odyssey-3 can now simulate general environments with far greater physical accuracy, all in real time. The knowledge learned by models like Odyssey-3 can also power robots, humanoids, driverless cars, and drones. And world models are just getting started! https://t.co/ZZDqKTdesK
The Hermes Webapp just landed on main! hermes update --channel main hermes config set webapp.enabled true hermes gateway restart hermes webapp It's local-only by default. For remote access, put it behind a sign-in. I run mine behind Tailscale with Nous login. Cheers~
Beautiful work from @perplexity_ai @AravSrinivas with weights on @huggingface of course! https://t.co/TJcVJlFYhg https://t.co/MJXAp5WUSt
Our next generation of open weights multimodal decision models comes in two sizes, pplx-decider-v1.2 and pplx-decider-flash-v1.2. Text and vision scores improve. Inference is up to 5x faster: all questions in a request run in one pass. Flash is the top 4B on the Decision Index.
๐ The latest @code release makes the Copilot SDK-based harness the default for new sessions, adds model comparisons, and brings a file list to multi-file diffs. ๐งต Hereโs whatโs new: https://t.co/teoiyQRELc
Yet again, AI development is moving faster than even the most aggressive predictions. With 3 days left in October, Google tied the human baseline on Blueprint-Bench 2. https://t.co/vH9XHDEhOR
Timelines just got shorter... AI will understand the physical 3D space better than humans by November* https://t.co/yRhjtVIMYM
Sakana AIใไบๅๅญฆ็ฟใใLLM้็บใใใใจใ็บ่กจใใพใใใๅ ใใฆใใฃใธใซใซAIใซๅใใใฏใผใซใใขใใซใ้็บใใพใใใพใใใฎ้็บใใฉใๅน็ใใๅ ้ใงใใใใAgentic Scientistใฎ็ ็ฉถใ่กใใพใใAI็ ็ฉถใฎๆๅ็ทใ่ตฐใใใผใ ใจๆฅๆฌใงไธ็ทใซๅใใใจใใงใใฆใใใใใใจใฆใๅฌใใใงใใ
Your AI coworkers now have an office. @getskydive is open to everyone. Build a team of agents, watch them work in a 3D office, and hand them tasks over Slack, email, iMessage or the web. Claim your companyโs domain to put your office on the map. They work for free this week. https://t.co/zqhQB0NvlV
Sakana AIใไบๅๅญฆ็ฟใใLLM้็บใใใใจใ็บ่กจใใพใใใๅ ใใฆใใฃใธใซใซAIใซๅใใใฏใผใซใใขใใซใ้็บใใพใใใพใใใฎ้็บใใฉใๅน็ใใๅ ้ใงใใใใAgentic Scientistใฎ็ ็ฉถใ่กใใพใใAI็ ็ฉถใฎๆๅ็ทใ่ตฐใใใผใ ใจๆฅๆฌใงไธ็ทใซๅใใใจใใงใใฆใใใใใใจใฆใๅฌใใใงใใ
ๆฌๆฅใฎSakana AIใฎใทใณใใธใฆใ ใฏใใใใใใพใงๆบๅกใจใชใใพใใใ่ถณใ้ใใงใใ ใใฃใ็ใใพใๆฌๅฝใซใใใใจใใใใใพใใ๏ผ๐ ใฆใซใฒใณใปใทใฅใใใใใผใใผๆๆใฎๅบ่ชฟ่ฌๆผใๅฏพ่ซใๅซใใใคใใณใใฎๆจกๆงใฏๅพๆฅใใใใใฆใๅ ฑๅใใพใใ ๅฝๆฅใฏ CEO ใฎ David Ha ใใใAIไธปๆจฉใซใจใฃใฆ้่ฆใจใชใ Sakana AIใฎๅใ็ตใฟใฎใ3ใคใฎๆฑใใซใคใใฆใ่ฉฑใใใพใใใ ใปPre-training๏ผๆฌกไธไปฃใฎไบๅๅญฆ็ฟใขใใซใฎ้็บใซAIใจใผใธใงใณใใๆดป็จใใ ใปPost-training๏ผๆฅๆฌ่ชใใใธใในใทใผใณใซLLMใ้ฉๅฟใใใไบๅพๅญฆ็ฟๆ่ก๏ผSak
Look friends. Here's what you need to know about "user instructions" for AI agents. What *you intend* is different from *what a system processes*. Saying "do X" does not rule out "don't do Y". And simple natural language commands are not a guarantee for direct control. 1/
In May, we had just closed our best quarter ever โ and of course decided to re-build @hexanalytics from scratch. Today we're announcing two things: Hex 4.0, the first analytics platform truly built for agents; and a $100m fundraise at a $1.2B valuation, led by @peakxvpartners The world is changing fast, and we're not slowing down! Read more here: https://t.co/FrhzVYu81f
Introducing Slingshots // FOUR: 14 projects across eight top institutions. Self-improving agent systems and harnesses, maximally-open models and training data, applications for research and decision-making โ plus kernels and more. Say hello to the batch ๐งต
Excited to share Halcyon, our latest research on multimodal pretraining! Building strong text-to-image generative models is simpler than we think! Blog: https://t.co/HOrwIJIHTf Models: https://t.co/xBPJkEAOPU Paper: https://t.co/8kROIz1c7L
1/8 Does text-to-image generation need language pretraining? Today, we share Halcyon, a family of early-fusion, base models trained from scratch. Halcyon outperforms strong late-fusion baselines with up to 275ร less pretraining compute.ย Models and recipes are open. Halcyon is p
Runway MCP is now in Grok Bot. Bring state-of-the-art generation into your chat and create anything you can imagine, without leaving Grok. Get started at the link below. https://t.co/FA6ugHOqDd
New on MicroDuck Arena: Tug of War ๐ฆ๐ชข๐ฆ Two robot ducks, one rope, a pit in the middle. Drag your rival in and you win. Train yours in simulation, publish it, and it's in the first cup, live Friday, Nov 6 at 9 AM PT. https://t.co/ZjmNwv89cV https://t.co/wYc1BqbN6t
I cannot express to you how much faster my ability to iterate on work became thanks to switching from Codex and Claude Code to a single unified Hermes Agent for Actual, and a single unified Hermes Agent for personal projects. It's like having a mind-reader! https://t.co/r3LuTIycLT
Modular has made the full MAX source available, including the compiler, runtime and hardware backends. Developers can write once, run across accelerators and build commercial products on MAX without licensing fees or royalties. A big step toward making heterogeneous computing more accessible across CPUs, GPUs and accelerators. https://t.co/YpVRIRW4yV
This entire game has zero asset files ๐ No textures No meshes No monster models No images No audio files The Ninth Vigil is a Diablo-inspired dark fantasy action RPG where everything you see and hear is generated in code when you press play And the crazier part? It was built by @NousResearch Hermes agents The world, enemies, combat, effects, sounds, UI, all of it Runs directly in your browser Free. Open source. I asked Hermes to build a game Apparently it decided it didnโt need an art folder ๐ https://t.co/HZLHQt6SYa
Hermes already writes code, makes images, does voice and runs its own sub agents Put all of that in one loop and you basically have a tiny game studio - One agent writes the game logic - One makes the sprites and art - One does music and sound - One plays the build and files th
Ornith-1.5-122B is now available for FREE in Nous Portal for 48 hours thank you @ornith_ If you are not yet, you will soon be tired of winning (but your Hermes Agent will never tire)
Aloha!๐บ Meet Ornith-1.5-122B: No Tradeoff Between Strong and Fast Stronger. Faster. Leaner. Ornith-1.5-122B delivers frontier-level capabilities with exceptional inference efficiency. With superior instruction following, stronger task decomposition, and deeper understanding of
captured more pencil traces and lined it up to the airpod traces next step is improving how the model handles orientation and speed using these recordings https://t.co/caKgTKkKpH
๐งต I've just released Sentence Transformers v6.2.0! Featuring model merging with @arcee_ai mergekit, load models faster, & evaluators share work across GPUs automatically. Plus lower-memory losses, multi-vector improvements & plenty of fixes. Thread with highlights ๐ https://t.co/V0HrkPiBpC
