All AI Pulse updates tagged "rohanpaul_ai".
119 signalsAnother example that the harness, more than the model itself, decides how far intelligence actually gets. With the API alone, execution ends when the model stops. @atomicagent_io , a model-agnostic agent layer, made GLM 5.3 better and it costs only 77 cents mo
Brett Adcock, founder of the U.S.-based Figure Robots, has a very strong opinion about Chinese robots. ---- From "My First Million" YouTube channel, (full video link in comment)
the drunken robot. at the world humanoid robot games in Beijing
What does it take for an AI to stay useful when a task lasts for hours—or weeks—and the world keeps changing? Goals evolve. New constraints appear. Plans fail. Meet dots3-note Preview, an open-weight multimodal model developed by rednote’s dots studio @dotsstu
Super interesting test by @atomic_chat_hq Shows that more quantization precision does not automatically produce the better result. Here, Qwen 3.8 27B Q8 (8-bit quantized version of the model) sometimes matched or beat Q4 (4-bit quantized) on the same voxel tas
This is bigger than a robot race. History just got rewritten. A humanoid robot from Beijing clocked 9.39s over 100m, faster than Usain Bolt’s 9.58s record from 2009. And this year's 100m was autonomous-only, i.e. the robot itself had to complete the race witho
the "do-not-touch" rule suddenly makes perfect technical sense 2026 World Robot Conference in Beijing shows an actual UBTECH ultra-bionic humanoid (semi-torso model) with highly realistic silicone skin ~ $17K and for companionship use.
The running part was pure mechanical poetry. An then that robot just rewrote the definition of quitting while winning.
So much great work is coming on the memory layer of AI. And this looks like super close to fixing the part of gmail i hate most. @getlindy 's new Chrome extension moves the AI agent assistant right into my Gmail itself. Removs the need to switch to a separate
She looks ready for a longer chat. AheadForm’s Elf Xuan 2.0 was exhibited at the World Robot Conference (WRC) 2026 in Beijing. The face alone packs 30+ degrees of freedom under half-mm silicone skin. powered by quiet brushless motors that keep eye contact and
This 3D coding test by @thehypedotnews found DeepSeek-V4-Pro-0813 used 48X more tokens than Muse Spark 1.2. thats why for agentic coding, the model's ability to reach an answer matters almost as much as the answer itself. A model that keeps reopening files, re
US VP JD Vance on AI Company's marketing strategy, some AI CEOs sell fear because fear makes their product look powerful. And his view on AI driven job loss. "I think the AI companies themselves, the CEOs, have a certain incentive to be super dystopian because
OpenAI President Greg Brockman just said OpenAI's roadmap points to major capability increases, as the model curve continues to advance. "We really have a roadmap lined up where the capabilities will continue to increase in significant ways that are hard to ki
Chinese open-weight models are gaining European traction. As it gives another route away from dependence on proprietary foreign APIs. Siemens has publicly described experimenting with Qwen and DeepSeek on a self-contained LLM platform that can run open-weight
AI policy gets distorted when forecasts pass as facts. “Bring science, not science fiction, back to the AI debate” - @drfeifei Asset managers face it when pricing AI into earnings and value. @ChinaAMC_HQ #ChinaAMC #CNQQ #AI4
Interesting choice of title from The Wall Street Journal. Nvidia and 6 financial firms have signed memorandums to mobilize more than $500B of third-party capital through independent compute-financing platforms. - Once Wall Street accepts GPUs as recoverable co
Soniox TTS v2 is one of those models you need to actually hear. They just launched TTS v2, a text-to-speech model that speaks 60+ languages from one mode. Really premium voice quality at a dramatically lower price ($ 0.70-per-generated-hour). > One model repla
Tailoring for machines is going to be a real job Una, UBTECH’s ultra-bionic humanoid, underwent makeup application and fabric adjustments as part of its first fashion modeling session.
The boundary between a specification and a product is dissolving. A Markdown file can now behave like a small software service when an AI agent can read instructions, use tools, and act on schedule. "Do you know how many companies are at this event where their
Meta's open-sourced Muse Glimmer 30B vs Muse Spark 1.2 Glimmer delivered 5 runnable games at roughly one-fifth Muse Spark's cost. Very Interesting experiment by @aimlapi . the setup: • Muse Glimmer 30B — via aimlapi[.]com. cost: $0.02 • Muse Spark 1.2 — via ai
I tested Uncensored[.]AI with questions that are sensitive enough to make most AI models become overly careful. It goes where no traditional chatbots will go. Will answer questions that all other popular ones would faint at the thought of answering. Forget pol
Microsoft reportedly plans to unveil Maia 300, its custom AI silicon. Also it wants TSMC capacity for 300,000 Maia 300 chips for future delivery. Microsoft says the chip delivers over 30% more tokens per dollar than the latest silicon in its fleet. More in-hou
Anthropic is shifting data center construction risk to infrastructure investors while keeping capacity through leases. Anthropic, Macquarie Asset Management and GIC are establishing Theseus Infrastructure, a platform that will develop, operate and lease purpos
Unitree Robotics' Shanghai IPO was roughly 5,526 times oversubscribed by retail investors. https:// x.com/UnitreeRobotic s/status/2069751801096909214/video/1 …
Vinod Khosla’s warning for India's BPO in the age AI: The traditional IT services and BPO business “will be gone” But India can still win if it shifts to deploying AI. ---- From "SparX by Mukesh Bansal" YouTube channel, (link in comment)
There's enough reason to be optimistic in this direction. We stopped accepting infection, we stopped accepting bad hearts, smallpox is gone, childbirth stopped killing women at scale Aging is just next on a very consistent list.
Great paper on Self-evolving agents. Enterprise agents cannot truly improve until their messy daily work becomes safe learning data. A future enterprise agent may improve by updating memory before changing its underlying model. The problem is that deployed age
Another reset for Codex and ChatGPT Work again 100% of weekly limits for paid users
So, OpenAI is also a signatory to this.
Rohan Paul @rohanpaul_ai · 16h Open Weights and American AI Leadership From microsoft.com 1 1.1K
Claude Opus 5 beat Claude Fable 5 on a hard 3D coding test at 3/4 the price. cool experiments by @thehypedotnews - opus 5 vs. fable 5 vs. gpt 5.6 sol vs. kimi k3 opus 5 produced the best-engineered and most visually convincing Three.js results wondering why an
The 1st open-source local agent to surpass Hermes on GAIA benchmark. Loop design, not model size, drove this whole gap. @atomicagent_io , a local-first AI agent that runs on your machine, with local or cloud models. Open source on macOS, Windows & Linux. Both
The hardware wave is picking up speed. Genspark just launched its first hardware, SecondBrain Note, a card-sized thin recorder pairing 35hrs of audio capture with searchable meeting memory. The 2.95mm, 26g aluminum card uses 4 microphones and 64GB storage for
BlackRock CEO Larry Fink on why china is ahead in the AI energy race. "We don't have the ability to build fast enough. And then I go watch China. China's building 100 gigawatts of nuclear. They're building close to 100 gigawatts of solar." https:// x.com/ArkkD
The $10-$15 trillion total addressable market for AI, if it is successful, is actually "terrifying". - The famous "Dean of Valuation", Professor Aswath Damodaran, of NYU Stern School of Business. The reason: AI as a tool is a much smaller market; AI as a repla
atomic[.]chat, a desktop app that runs LLMs locally, ran a very revealing comparison for Claude Sonnet 5, Claude Opus 4.8, Claude Sonnet 4.6, and GPT 5.5. Claude Sonnet 5 just matched GPT 5.5 on 3 physics coding demos at 6x lower cost. Also spent minimum numbe
Meta open-sourced a brain-to-text system that reaches 78% word accuracy without surgery. Brain2Qwerty v2 converts non-invasive brain recordings into text with 61% average word accuracy and 78% for its strongest participant. The system reads MEG signals from a
Most AI products ask users to leave their workflow and enter a separate box of intelligence. Acti ( @openacti1 ) reverses that direction by putting the agent inside the text field, where plans, questions, replies, reminders, links, and decisions already begin.
"If we could snap our fingers and get a pile of data... we would solve general robotics right now." - Figure CEO Brett Adcock The big bottleneck in Physical AI / robotics is not better models, but better robotics data infrastructure. That is the gap @cyberorig
FT: Google capped Meta’s use of Gemini after Meta asked for more model compute capacity than Google could supply. Meta’s problem is that it uses Gemini inside safety automation, customer support, ad tools, coding, and internal workflows. Google’s problem is di
Reuters: Austria is asking the EU to give Anthropic a European base because US AI controls have turned frontier-model access into a geopolitical dependency. It is asking Brussels to explore a political and legal package that could make Anthropic run part of it
Box's CEO discusses AI access as a new tool in geopolitics.
The State of the AI Economy shows $110B real AI revenue over 12 months.
Microsoft just turned Copilot in Excel into a finance workflow system So it can follow team-defined methods, pull institutional data, and leave an audit trail for every workbook edit. The new connectors bring finance data from FactSet, Morningstar, PitchBook,
This act requires AI data center builders to pay for new power plants and upgrades.
OpenAI released a paper showing agents executing most tasks in office work.
Non-developer Codex adoption is up 189x in organizations and 137x among individuals.
OpenAI is becoming a test case for workplaces where humans supervise agents.
i10X just introduced Superagent, an AI Chief of Staff built to turn business goals into finished work. By giving an AI agent access to 100+ tools, cloud execution, research, writing, prospecting, decks, SEO, and human approval gates. Most AI tools still depend
AI agents could mediate $3 Tn to $5 Tn in consumer commerce by 2030, transforming retail shopping.
Raoul Pal explains how agentic AI can replicate and optimize software on demand.
This diagram shows the paper’s idea that intelligence can be modeled as a small meaning-based decision map, where clues like danger, food, urgency, and location push the system toward 1 of 4 actions: hide, escape, stay still, or move slowly. The point is that
Intelligence may relate more to better knowledge structures than larger models.
Brilliant NVIDIA’s Rubin AI servers can now cool every chip and networking part with 45°C water-glycol coolant instead of cold air. The big deal with this is that cooling water use can drop from about 2.6M gallons per MW per year to near zero in suitable clima
Nice. Getting this same feeling around, a quieter shift in AI coding: powerful agents moving back onto private machines. curious how far this can go with mid-range consumer hardware.
Atomic Chat just made Cline run coding agents on local AI models. @cline is the agent layer: it can read files, edit code, run terminal commands, inspect errors, and ask approval before actions. Atomic Chat is the model layer: it runs open-weight models offlin
The self-resampling part of MaineCoon-22B is crucial for training.
Link to the technical blog of MaineCoon-22B and apply for access.
AI video is moving into real-time reaction era, with MaineCoon leading.
MaineCoon preserves human presence across time, key for social world model.
Funding sources include $1.4T from hyperscalers and more.
History tells us there may be a bust in the future.
Proposed closer work with Trump administration for security.
Nypost: Anthropic is trying to get Washington to reverse the US block on its most powerful Mythos Anthropic has proposed working more closely with the Trump administration, improving communication, and resolving security concerns faster as it seeks to end the
The kill-switch debate received user feedback.
Viktor grew a $20M annualized revenue run rate outside Microsoft Teams. Now it works inside Teams. This revenue is from an AI employee that does the job, not one that just replies. Try free at @viktor__com . $100 in credits, no card.
And the Github of Atomic Mail https:// github.com/Atomic-Mail/at omic-mail-agentic …
Yann LeCun ( @ylecun ) explains why LLMs are limited in terms of real-world intelligence during a Bloomberg interview. "Language is a very approximate, reduced, quantized, and simplified description of the world, and LLMs can only deal with discrete sequences
Agents can now have their own email! @atomic_mail just launched something to fix a missing piece in agentic workflows: agents need inboxes of their own, not borrowed human ones. So Atomic Mail connects to popular agents like Claude Desktop, Cursor, OpenAI-base
Try it here now https:// atomicmail.io/agents
Noam Shazeer, a key model builder, is moving to OpenAI.
He discusses the commercial losses from not releasing the model.
Anthropic just rolled out a major Claude Design update, adding design system imports and code round-trips.
The G7 AI lunch showed that frontier AI leaders are part of geopolitical decision-making.
Genspark's AgentBase allows users to turn existing data into internal tools.
Logarithmic AI compute chips are 17x more tokens per watt.
This is their more detailed technical report.
Microsoft’s 2026 Work Trend Index shows Singapore workforce ahead on AI adoption. 66% of Singapore AI users say they now produce work they could not have made a year ago, compared with 58% globally. Among Frontier Professionals, Microsoft’s term for advanced A
AI diffusion over time by economy
The KV cache management is probably doing more work here, than most people will notice. Continuity lives in memory, not only in pixels. Great project.
Check out this paper on AI.
Pythagoras-Prover shows Lean theorem proving is less dependent on giant models.
Connects AI agents to the entire software workflow.
This article is from the CEO of @FactoryAI.
Especially crucial for long-running agents.
There is so much potential in robotics training data for the next trillion-dollar industry.
High-quality interactions can train weaker student models to imitate useful behaviors.
Great article by Satya Nadella on organizational economics of AI and 'token capital', emphasizing the importance of learning loops.
This is why generic model performance is only half the story for enterprise AI.
This is why generic model performance is only half the story for enterprise AI.
Traditional IT services and BPO will be gone, but India can win with AI.
Current AI agents struggle to pass real human verification checks.
AgentPerf treats agentic AI as a long-running workload, considering memory and networking.
NVIDIA posts benchmark results showing GB300 NVL72 runs 20x more coding agents per megawatt.
Opportunity of the century.
The scary part is what happens when AI learns the last 10%.