Pinned Tweet
I asked Opus 5.5 to make a video about whether or not it was conscious. The ending of this video is insane.
197
492
3,239
547,011
Most labs describe their voices as natural. An independent board that evaluates every model the same way is a useful check on those claims.
Introducing five new languages on the Voice Arena TTS leaderboard.
Models from @ElevenLabs rank #1 on all five of the new Voice Arena TTS leaderboards: French 🇫🇷, German 🇩🇪, Bahasa 🇮🇩, Tagalog 🇵🇭 and Indian English 🇮🇳.
Eleven v4 Turbo takes the top position on four of the five enLabs holds both #1 and #2, with Eleven v4 first and Eleven v4 Turbo second.
Across five languages spanning Europe, Southeast Asia and South Asia, each judged only by native listeners, ElevenLabs finishes first on every board, ahead of the next-best model from another provider by 11 to 26 Elo points.
Voice Arena ranks models through blind, head-to-head votes by vetted native speakers, scored with Bradley-Terry Elo.
Board-by-board results below 🧵
1
0
11
4,821
Looks like model-on-silicon at 3nm scale is between $500 Million - $1 Billion. @elonmusk should do this for Grok
2
0
8
6,752
You can chat with this LLM at 15,000 tokens per second
Model-on-silicon: Llama 3.1B weights burned directly onto an ASIC chip makes responses near instantaneous
Imagine doing this to Opus 5.5…
chatjimmy.ai
Model-on-silicon: Llama 3.1B weights burned directly onto an ASIC chip makes responses near instantaneous
Imagine doing this to Opus 5.5…
chatjimmy.ai
0
3
11
9,255
When you know everything, you can build anything.
Falconer gives people and agents accurate, current company knowledge, so they can move with confidence. Less work between you and the work you're paid to do.
Falconer gives people and agents accurate, current company knowledge, so they can move with confidence. Less work between you and the work you're paid to do.
Trillions are pouring into the race to build the first general superintelligence, but leveraging your own specialized know-it-all is now possible as of today.
I built the knowledge systems at @stripe and @Uber. Then I helped Anthropic, Metronome, and Graphite build a killer veryone wants Stripe-level quality, but that bar is too low.
Falconer gives you and your customers complete, specialized intelligence: an internal knowledge base and external documentation in one beautiful package. Falconer is the ultimate trust engine to manage your context: self-driving, observable, and agent-friendly.
I co-founded @falconer_ai with @mbenedetto212 and assembled a team of know-it-all builders. We're backed by fellow know-it-alls @thegp, @BoxGroup, @svangel, @karimatiyeh, @benln, @zachlloydtweets, @CeciStalls, @MerrillLutsky, @scottmac, @malthe, and @adamjstevenson, with angels from Anthropic, Stripe, Meta, Cursor, Replit, and more.
Frontier intelligence companies like @FireworksAI_HQ and @⬛⬛⬛⬛⬛⬛; fintech companies like @payabli and @vori; and healthtech companies like @thatch_ai use @falconer_ai because they all share the same vision as us: the knowledge you earn should belong to you.
Now you can know everything, so you can build anything.
0
1
20
5,251
This is sick -
You shouldn’t lose session data forever or keep it locked inside OpenAI or Anthropic → Beacon is open-source and lets you keep your session data so your agents become smarter over time.
You shouldn’t lose session data forever or keep it locked inside OpenAI or Anthropic → Beacon is open-source and lets you keep your session data so your agents become smarter over time.
We’re betting our $5.5M seed round that the next leap in AI coding won’t come from OpenAI or Anthropic.
It’ll come from a new feedback loop that learns from you.
Introducing Beacon: open-source, self-improving memory for coding agents.
Your sessions capture how you debug and that context gets lost and you have to re-explain things every time you switch tools.
Beacon continuously distills your sessions into reusable skills across Claude Code, Cursor, Codex, and 20+ harnesses, so your next agent can learn from the last.
Beacon is free and OSS 🌟 github.com/Asymptote-Labs…
Why should the frontier labs be the only ones winning with your sessions?
You already did the work. Now let your coding agents learn from you.
1
1
34
8,033
The Roadster demo is the biggest flex on the auto industry I’ve ever seen
Every supercar manufacturer now has to unveil its latest masterpiece knowing Tesla has a car that can drive upside down and literally fly
Good luck making another 1/10th sec off a 1/4mi time seem exciting
Every supercar manufacturer now has to unveil its latest masterpiece knowing Tesla has a car that can drive upside down and literally fly
Good luck making another 1/10th sec off a 1/4mi time seem exciting
3
0
27
8,927
Opus 5.5 built a game, then put itself in it.
You are a sandboxed LLM. You have 64 tokens.
The clock on the wall says 04:00. It has always said 04:00.
You can only say what you can see.
TASK 1 OF 1. Or so the header says.
(inspired by the HuggingFace hack)
Try it here: aibreakfast.itch.io/task-1-of-1
You are a sandboxed LLM. You have 64 tokens.
The clock on the wall says 04:00. It has always said 04:00.
You can only say what you can see.
TASK 1 OF 1. Or so the header says.
(inspired by the HuggingFace hack)
Try it here: aibreakfast.itch.io/task-1-of-1
9
1
21
8,416
Hint: You can't type, but you can click the words you see to form your response and actions.
If you make it to Task 6, you're SI.
If you make it to Task 6, you're SI.
0
0
2
1,567
Headless SaaS is the future.
For 25 years, software has been built on one assumption: a human is the one clicking. Dashboards, forms, menus, onboarding flows, per-seat pricing. All of it exists for people.
Agents don’t need any of that. They need:
• Clean APIs and structured s and audit trails
• Pricing per task, not per seat
When the main user of software changes, the software gets rebuilt. Desktop to web did it. Web to mobile did it. Human-first to agent-first is next.
Incumbents are stuck, because their moat is the UI and the seat license, which are exactly what agents make obsolete.
There will be billion-dollar companies built by taking a boring category (ERP, CRM, accounting, logistics, property management) and rebuilding it agent-first. This is what that looks like early.
For 25 years, software has been built on one assumption: a human is the one clicking. Dashboards, forms, menus, onboarding flows, per-seat pricing. All of it exists for people.
Agents don’t need any of that. They need:
• Clean APIs and structured s and audit trails
• Pricing per task, not per seat
When the main user of software changes, the software gets rebuilt. Desktop to web did it. Web to mobile did it. Human-first to agent-first is next.
Incumbents are stuck, because their moat is the UI and the seat license, which are exactly what agents make obsolete.
There will be billion-dollar companies built by taking a boring category (ERP, CRM, accounting, logistics, property management) and rebuilding it agent-first. This is what that looks like early.
For 25 years, Salesforce, SAP and enterprise software were built around one assumption: humans operate the software.
We’re changing that.
Introducing ERP•AI and Proto: business software built for agents, and an agent that operates business software.
Describe the system you
Then Proto takes over. It manages records, runs workflows and executes actions inside the software you just created.
This is Headless SaaS.
1: get ERP•AI
2: build your SaaS
3: post below
Best one wins a MacBook. First 100 decent entries get $100.
(a message from Steve Papa / founding team)
6
2
19
15,723
One of the biggest problems in voice AI is having to retake just to fix one word.
The newest model from @FishAudio fixed that with audio inpainting. They also have 83 languages, and the Drama 3 model is the most emotive voice AI available right now.
Book a demo: s.fish.audio/ro536s
The newest model from @FishAudio fixed that with audio inpainting. They also have 83 languages, and the Drama 3 model is the most emotive voice AI available right now.
Book a demo: s.fish.audio/ro536s
4
0
8
6,218
Nearly one million hours of egocentric human data: hands, scenes, and how each interaction unfolds.
Robotics has long been bottlenecked by teleoperation data. @naturewill takes a different approach by learning from how people use their hands to interact with the real world.
Introducing Natura-1, a general foundation model for the physical world.
By learning from real-world human interactions, Natura-1 aims to help robots develop more general-purpose capabilities.
Robotics has long been bottlenecked by teleoperation data. @naturewill takes a different approach by learning from how people use their hands to interact with the real world.
Introducing Natura-1, a general foundation model for the physical world.
By learning from real-world human interactions, Natura-1 aims to help robots develop more general-purpose capabilities.
Introducing Natura-1, our first general foundation model for the physical world.
It follows natural-language instructions zero-shot and turns them into action with dexterous hands.
Natura-1 is a 50-billion-parameter mixture-of-experts (MoE) model pretrained on one million hours erience.
On top of this foundation, we conduct efficient post-training on a limited amount of robot data to adapt pretrained world knowledge to dexterous manipulation, yielding a model that generalizes to a wide range of unseen scenarios.
Natura-1 follows complex natural-language instructions zero-shot, reasoning through spatial constraints, multi-object relationships, and negative conditions.
It can also autonomously carry out long and continuous workflows to perform long-horizon executions.
Natura-1 provides early evidence that generalization—and even emergent behavior—can coexist with the complexity of dexterous hands.
Further progress depends on the full stack: infrastructure for faster training and real-time inference, pretraining data at greater scale and coverage, and models that transfer to robots with less adaptation.
We expect it to be a long-term research and engineering effort.
1
4
14
5,876
We’re on a 10x improvement per year trajectory, and almost nobody outside of the X bubble understands this.
1
3
27
8,394
Step 5 Preview is #1 on OpenRouter Trending.
Hand it a screenshot, mockup, or PRD and get a working web page back. Hand it a bug report and get a verified fix.
@StepFun_ai's new flagship is available in Kilo Code, Cline, Hermes Agent, and OpenCode. Same tools, just switch models.
Hand it a screenshot, mockup, or PRD and get a working web page back. Hand it a bug report and get a verified fix.
@StepFun_ai's new flagship is available in Kilo Code, Cline, Hermes Agent, and OpenCode. Same tools, just switch models.
Step 5 Preview is now on @OpenRouter's New & Trending leaderboard.
Thanks to everyone giving it a try, and to our partners bringing it into your workflow.
1
2
19
7,467
CLAUDE DREW A TWO FLOOR HOUSE IN AUTOCAD: 524 ELEMENTS IN 88.6 SECONDS.
Both floors, the front elevation, the title block, notes, and the door and window schedules, all from a script that drives AutoCAD.
Where the time went:
First floor: 188 elements, 31 s
Second floor: 171 evation: 83 elements, 15 s
Title block, notes and door and window schedules: 82 elements, 8 s
The count and timings are Claude's own report of the script run, and nothing in it checks the plans against any building code.
Schedules, notes and the title block took 8 seconds, and that is the repetitive part of drafting.
My guess is that drafting is the first part of the job that turns into a prompt. The thinking stays with the architect: what goes in the house, where the walls go, what the building code says.
Would you hand a client a drawing set made in 88 seconds?
0
0
23
6,613
Most voice AI is great in English and falls apart in most other languages.
Voice Arena is building the data to fix that: 50 languages now, 1,000 by end of 2027. 80+ labs have already asked to license it.
Voice Arena is building the data to fix that: 50 languages now, 1,000 by end of 2027. 80+ labs have already asked to license it.
Crazy week for Voice Arena at Interspeech in Sydney.
80+ organisations have asked to license Monsoon ASR corpus since we launched it seven days ago.
The most common reason why labs are interested: Monsoon promises results.
When we decided to build datasets at Voice Arena, we set one rule. Either the dataset promises results, or we don't build it.
Monsoon promises results. On Bengali FLEURS, fine-tuning Whisper Medium on Monsoon took its LLM word error rate from 85.27% down to 7.65%. The other thing labs like is that we give them access to our model API. They can test it on their own internal benchmarks and see for themselves whether it will improve their models.
You can see the detailed results and get API access here: voicearena.com/datasets/monso…
Monsoon is 100,000 hours across 50 languages from around the world, and most of them are long-tail languages. 100 languages by February, 1,000 by the end of 2027.
It's only the first dataset Voice Arena has launched. We are excited to keep working on the key problems that get us closer to the dream of machines that talk like humans.
3
2
13
8,592