Erica Sandberg 舊金山的神奇女俠
TISK TISK! Ripping down Saikat Chakrabarti/Hasan Piker protest signs is not cool. Know this lovely lady? DM me. Oh, and the rally is Thursday, 715 Harrison St 5:00.
Claude Code does sometimes make me laugh out loud
Browser Use
Hermes agent just gained a new skill: browser-harness
Now, Hermes agent has:
> self-improving browser tools
> parallel stealth cloud browsers
> full freedom within your browser
All it takes is one prompt.
Try it now ↓🔗
ChatGPT feels very 'switched on' now
Sam D'Amico
Continuous-video agents (computer use, robotics, static scenes) burn compute re-ingesting pixels that didn't move.
VLMaxxing teaches a frozen video VLM to skip the reruns. 54 fps perception on Gemma 4 26B, training-free, no accuracy drift.
w/ @jfbastien (arXiv 2605.03351)
A cybersecurity firm, “Red Access,” contacted us less than 24 hours before going to the media with vague claims about Replit.
This is not how responsible security research works. The standard practice in terms of disclosure policies, as followed by CISA, CERT/CC, and most major firms, is to share findings privately and allow a defined window for remediation before public disclosure. A 24-hour countdown to a press cycle is not that.
From the limited information they shared, their core claim appears to be that some users have published apps on the open web that should’ve been private.
Replit allows users to choose whether apps are public or private. Public apps being accessible on the internet is expected behavior. Privacy settings can be changed at any time with a single click.
Vibe Coding is a rapidly developing space, and we take our responsibility to both provide tools to create secure apps and educate our customers very seriously. Just in the past week, we launched two security products: Security Agent and Auto-Protect.
If Red Access shares a list of impacted users, we will proactively default those apps to private and notify users directly.
We welcome responsible security research and have a long history of working constructively with researchers who follow standard disclosure practices. That offer remains open.
Eugene Yan
some thoughts on working with ai models
• context as infra
• taste as config
• verification for autonomy
• scaling via delegation
• closing the loop
https://eugeneyan.com/writing/working-with-ai/
ChatGPT for Excel and Google Sheets:
ChatGPT: ChatGPT is now available as an add-on in Excel and Google Sheets.
It can help analyze messy data, write formulas, update spreadsheets, and explain what it’s doing along the way—without leaving your spreadsheet.
Powered by GPT-5.5.
https://chatgpt.com/apps/spreadsheets/
typical @evanjconrad W
Andrew Curran: BlackRock CEO Larry Fink predicts a compute futures market.
Prop D is a smoke screen by dishonest people who claim to want to hurt overpaid CEOs but in actuality are going to get retail closed in SF while passing an 800% gross receipts tax that will be paid by the middle and working class
Blueprint: Supervisor @bilalmahmood pitches Prop D as a Medi-Cal and CalFresh lifeline.
Read the measure. Zero binding language requiring a dollar go to health programs.
It all lands in the General Fund. That means pay raises, pet projects, whatever politicians want.
Sheel Mohnot
Both the democratic frontrunners for governor want to kill more people by reversing the autonomous vehicle rules that the CA DMV already approved
Tom Steyer: AI shouldn't put California truckers out of work to pad Big Tech's profits. As governor, I'll reverse the DMV's autonomous trucking rules and keep human drivers on the road.https://www.politico.com/news/2026/03/03/california-politics-could-cause-a-reversal-on-autonomous-trucks-00811162
GBrain v0.27.1 now supports multi-modal image embeddings
RFKjr == Lyssenko
The New York Times: Breaking News: The FDA has blocked publication of research that found widely used Covid-19 and shingles vaccines were safe. https://nyti.ms/49dtF24
Steven Rattner
President Biden made health insurance more affordable for millions of Americans in the ACA marketplace.
The Republicans’ actions increased prices by 58% on average—amounting to huge extra costs for the average consumer each year.
My @Morning_Joe Chart.
Marco Lobo
we run OpenClaw qmd as the memory layer across every agent we operate since the beginning of the year
benchmarked it against @garrytan gbrain. 150 real questions, our own corpus.
qmd lost 58 to 7.
then the data got more interesting. the part everyone markets did nothing.
JCat
The second half of the AI Agent competition is no longer just about who is more powerful; more importantly, it hinges on who understands users better.
We've compared the memory mechanisms of OpenClaw and Hermes, the self-evolving graph memory represented by gbrain @garrytan, and the precise retrieval memory mechanism exemplified by Mem Palace @bensig @MillaJovovich. Each of these three approaches has its own focus, yet they complement one another perfectly.
Yours aims to integrate these three mechanisms to endow AI with human-like memory, enabling it to grow alongside users.
JCat: http://x.com/i/article/2051975174883528704
Ryan Petersen
The palisades fire being an act of left wing terrorism is not being discussed nearly enough.
Drew Pavlou 🇦🇺🇺🇸🇺🇦🇹🇼: Palisades Fire ends up being one of the most destructive acts of left wing terrorism in modern history.
The arsonist Rinderknecht searched “free Luigi Mangione,” “let’s take down all the billionaires” and “let’s kill all the billionaires” before setting the Lachman Fire which
Lulu is like "The Wolf" from Pulp Fiction
Ti Morse: My first interview with @lulumeservey, Founder of Rostra.
0:07 How to Destroy a Terrorist Group
3:01 What Makes a Great Cult Leader
4:42 Unleashing Palmer Luckey
7:34 Why Elon Is Unpredictable
10:41 Demanding a Hardcore Culture After the X Acquisition
13:41 How Napoleon Rallied
Steven Rattner
Over 10 million Americans will lose health insurance because of Trump’s cuts.
4 million from losing subsidies and 6 million more from cuts to Medicaid.
My @Morning_Joe Chart.
Steven Rattner
Red states have disproportionately benefited from the health insurance subsidies expanded by Biden and then cut by Trump.
Georgia is a clear example: in just one year, marketplace enrollment has dropped by more than 550,000 people — a 37% decline.
My @Morning_Joe Chart.
Just in time software is here
Alfred Lin: Every era of computing invented its own way of building software. Waterfall optimized for reliability and control. Agile optimized for velocity and modularity. The AI era will optimize for direction and leverage, but we don't yet know what to call it. Who's going to define how we
Hubert Banville
🧠 Introducing NeuralBench: a unified, open-source framework to benchmark NeuroAI models.
v1.0: 36 EEG tasks, 94 datasets, task-specific + foundation models. MEG/fMRI ready.
MIT-licensed, FAIR's Brain & AI @AIatMeta.
Code: https://github.com/facebookresearch/neuroai/tree/main/neuralbench-repo
Paper: https://ai.meta.com/research/publications/neuralbench-a-unifying-framework-to-benchmark-neuroai-models/
Peter Yang
I've spent way too long testing OpenClaw, Hermes, Claude Code, Codex, and Gemini as my personal agent.
The truth is, nobody has won this race yet.
Here's my new deep dive with my honest take on where each product stands, plus the personal agent stack I use right now.
📌 Read now: https://creatoreconomy.so/p/the-race-to-build-a-personal-ai-agent-openclaw-hermes-claude-codex-gemini
Max Raskin
I talked to SF Mayor @DanielLurie about how he's bringing down crime dramatically in the city...which is making @realDonaldTrump less likely to send in the troops.
My latest in the @washingtonpost:
Todd Hanford
The wave of corporate reorgs which cut management layers gives a new meaning to "fat skills, thin harness" (TM) @garrytan .
The corporate equivalent:
Fat ICs, thin managers
What's true for agents is also true for humans, less constraints around strong workers allows them to express their talents in more productive ways. Additionally skill compression - i.e. one person teams - reduces the lossyness of communication - i.e. exec -> PM -> designer -> engineer.
This is actually a great trend for strong ICs. Traditionally the only way to make more money was to move into management. In the new agentic world, ICs are potentially more valuable than managers.
Brian Armstrong: This is an email I sent earlier today to all employees at Coinbase:
Team,
Today I’ve made the difficult decision to reduce the size of Coinbase by ~14%. I want to walk you through why we're doing this now, what it means for those affected, and how this positions us for the
Jean-Rémi King
💫Very happy to release NeuralBench, to benchmark Neuro AI models and datasets in the open!
🧵Thread, 💻Code, 📝White Paper below:
Hubert Banville: 🧠 Introducing NeuralBench: a unified, open-source framework to benchmark NeuroAI models.
v1.0: 36 EEG tasks, 94 datasets, task-specific + foundation models. MEG/fMRI ready.
MIT-licensed, FAIR's Brain & AI @AIatMeta.
Code: https://github.com/facebookresearch/neuroai/tree/main/neuralbench-repo
Paper: https://ai.meta.com/research/publications/neuralbench-a-unifying-framework-to-benchmark-neuroai-models/
Claude down?
At the Lab, the real excitement begins when our experiments reach your hands and we get to see what you create.
That’s why we’re super excited to share that @GoogleFlowMusic is partnering with Believe, a local artist development company, to put Flow Music directly into the hands of artists to help them explore, craft, and share music.
Straight music to our ears. Learn more in the post below!
Google Flow Music: Announcing a new partnership between Google Flow Music and global artist development company, Believe! 🤝 Here's what this means:
✨ Access to premier tools: Flow Music is now available to help all Believe and @TuneCore artists craft original tracks, ready for release.
🎙️
ClaudeDevs
Code with Claude is happening now!
▪︎ 9:00AM - Keynote
▪︎ 10:30AM - What's new in Claude Code
▪︎ 11:15AM - Building on Claude at GitHub scale
▪︎ 12:00PM - Get to production faster with Managed Agents
All times PT. https://x.com/i/broadcasts/1qGoNegbnRNKv
Kristian Ranta
I lost my brother Peter to suicide. Standard psychiatry failed him.
Today we launch Meru Health Advanced — the care model I wish he'd had.
For the next Peter.
Multipath Reliable Connection (MRC): a new open networking protocol for large AI training clusters, deployed in production on our largest training clusters.
OpenAI: We’ve partnered with @AMD, @Broadcom, @Intel, @Microsoft, and @NVIDIA, to release Multipath Reliable Connection (MRC), a new open networking protocol that helps large AI training clusters run faster and more reliably, with less wasted GPU time.
https://openai.com/index/mrc-supercomputer-networking/
NYU Center for Data Science
AI agents often struggle to plan movements because their internal representations of the physical world can be overly tangled.
CDS PhD student Ying Wang (@yingwww_) shows how straightening these pathways improves AI navigation.
Accepted to ICML 2026.
https://nyudatascience.medium.com/improving-world-models-a-neuroscience-inspired-approach-to-latent-planning-c745f493a5c2
1/
NYU Center for Data Science
Re Research done with NYU postdoctoral researcher @oumaymabounou, Courant PhD student Gaoyue (Kathy) Zhou, Brown University Asst. Prof. @randall_balestr, former CDS Faculty Fellow & U. Toronto Asst. Prof. @timrudner, CDS founding director @ylecun, & CDS Asst. Prof. @mengyer.
2/2
Latent.Space
experiment:
livetweeting the @AnthropicAI code with claude event!
first up - @katelyn_lesse and @angjiang on claude platform!
RT Armstrong and Getty
More chips, more Claude
Claude: Effective today, we are:
1) Doubling Claude Code’s 5-hour rate limits for Pro, Max, and Team plans;
2) Removing the peak hours limit reduction on Claude Code for Pro and Max plans; and
3) Substantially raising our API rate limits for Opus models.
I'll be at Code with Claude all day today so come find me and let's chat about Claude!
I'll also be giving a talk on the main stage at 530pm PT so tune in, it will be on the livestream!
ClaudeDevs: Code with Claude is happening now!
▪︎ 9:00AM - Keynote
▪︎ 10:30AM - What's new in Claude Code
▪︎ 11:15AM - Building on Claude at GitHub scale
▪︎ 12:00PM - Get to production faster with Managed Agents
All times PT. https://x.com/i/broadcasts/1qGoNegbnRNKv
Replit in Ghana!
Kweku Tech: Last Saturday, 40+ developers walked into Buro Labone with one challenge:
build something real in 4 hours using Replit Agents.
They walked out having shipped 20+ working products.
Here’s what happened at Build with Agent 4 by @ReplitGhana 🧵
#KwekuTechShotIT
Latent.Space
Re omg @bcherny with banger quotes
“the future is more async agents… this is why we emphasize verification”
“if you’re familiar with higher order functions, routines are higher order prompts”
“default is i will now have claude prompt claude code”
“the capability is already here - the gap left is how to put it to work.”
Elon Musk
Re Same here.
By way of background for those who care, I spent a lot of time last week with senior members of the Anthropic team to understand what they do to ensure Claude is good for humanity and was impressed.
Everyone I met was highly competent and cared a great deal about doing the right thing. No one set off my evil detector. So long as they engage in critical self-examination, Claude will probably be good.
After that, I was ok leasing Colossus 1 to Anthropic, as SpaceXAI had already moved training to Colossus 2.
Latent.Space
Re @swyx @maheshmurag deep dive on the new dreaming feature in beta today
https://x.com/claudeai/status/2052067400690851842?s=46
Claude: Dreaming reviews your agent's past sessions, extracts patterns, and curates memories so your agents learn over time.
Request access: https://claude.com/form/claude-managed-agents
clem 🤗
This week made something clear: you shouldn't take what most tech ceos are saying publicly seriously!
Yuchen Jin: From “Anthropic is Misanthropic” to “Claude is good for humanity and was impressed.”
Most ironic outcome is most likely.
Micah Erfan
Every time Republicans take power, they increase the deficit.
Every time Democrats take power, they reduce it.
Jake Sherman: CBO says the Senate Republicans' new reconciliation bill will increase deficits by $71.7B over the next 10 years. That's what the bill costs, so that makes sense.
https://www.cbo.gov/system/files/2026-05/HSGAC-and-Judiciary-Reconciliation.pdf
Ryan Delk
Fun to flip the script and interview my friend @amasad last week.
Replit is still underrated.
Join us at 1pm PT for a conversation with our co-founders Dario Amodei and Daniela Amodei, moderated by Chief Product Officer Ami Vora.
https://x.com/i/broadcasts/1qGoNegbnRNKv
Lenny Rachitsky
Announcing the Lenny's Newsletter x @Replit Buildathon
Build something awesome with my podcast and newsletter data using Replit, and win fabulous prizes:
🔸 1:1 career coaching session with @amasad
🔸 $5,000 in free Replit credits
🔸 A free year of Lenny's Newsletter
This is the excuse you've been waiting for to build with AI.
Learn more and enter the thread below.
P.S. The first 200 joiners get a free month of Replit Core.
Steven Rattner
Over 10 million Americans will lose health care coverage because of Republican cuts to Medicaid and Obamacare subsidies.
That amounts to half of the total number of people who gained health insurance because of the A.C.A.
📸: @Morning_Joe @MSNOWNews
This is exactly why we build. 🥹 Watching an experiment move into the hands of a maker’s creative process is what keeps us going.
It’s great to see how @PomellibyGoogle was able to help Michelle bring this beautiful collection to life. Huge congrats to the Inner Child team!
Google Small Business: How do you launch 60 lampshades when you’re already wearing every hat?
For Michelle of Inner Child, @pomellibygoogle from @googlelabs helped design the images by putting the lampshades on lamp bases to complete her latest collection ✨
Learn more at https://goo.gle/4ncpgT3
Dario: “We saw 80x growth earlier this year on usage and revenue”
“That’s why we are going to keep acquiring as much compute as we can”
Some nice quotes from Dario and Daniela from this session:
“I’ve been training these models since 2015, they were really dumb.”
"Build for the exponential. There are products that are not possible with the current model but could work with later models. This puts a premium on internal experimentation. You have to just try it with the model to see if it works."
“Products reach their saturation when models get too good. The ways in which we are making models smarter today are much more evident in agentic and Claude Code workflows than chatbots. You have to think about what the next big thing is.”
“Models are evolving from coding to software engineering to just growing a business.”
“It’s easy to accumulate an extraordinary amount of internal tech debt when you ship that fast. But the models can also help you pay down this debt."
Peter Yang: Dario: “We saw 80x growth earlier this year on usage and revenue”
“That’s why we are going to keep acquiring as much compute as we can”
Corruption, part N
No Lie with Brian Tyler Cohen: Palm Beach International Airport is now officially renamed to Donald J. Trump International Airport.
The airport will be required to purchase official Trump merchandise, directly violating the Emoluments Clause of the Constitution.
Corruption, part N+1
Republicans against Trump: Someone quietly placed a massive $920 million crude oil short at 3:40 a.m. ET this morning.
Just 70 minutes later, Axios reported the U.S. and Iran were close to a 14 point deal to end the war. Oil prices crashed, and someone made a fortune.
The golden age of corruption.
David Lieb
Re @AgentPhoneHQ is hosting Call My Agent Hackathon at YC on May 17th. Build agents that actually touch the real world across voice, SMS, iMessage, and more. No more sandbox demos. Over $10k in prizes and first place gets a guaranteed YC interview.
https://events.ycombinator.com/CallMyAgentHackathon
Y Combinator
Blake Scholl is speaking at Startup School 2026.
@bscholl founded Boom Supersonic with a goal of bringing back commercial supersonic flight. And in 2025, XB-1 broke the sound barrier — the first privately developed supersonic jet to do so.
http://ycombinator.com/startupschool
codex is for everyone
Marina Mogilko: .@thsottiaux told me on my podcast this week: more than half of Codex prompts now come from non-engineers
As a knowledge worker, I can't be more excited about what's shipping. Testing Codex this weekend. Will report back
Never has the 🚀 emoji felt more apt.
Tom Brown: In the next few days we'll be ramping up Claude inference on Colossus.
Grateful to be partnering with SpaceX here. We are going to need to move a lot of atoms in order to keep up with AI demand, and there's nobody better at quickly moving atoms (on or off planet Earth)
Replit ⠕
We're doubling down on keeping your Apps secure 🔒
Starting today, all Replit builders- free and paid- can publish their apps privately, making them accessible only to the people you choose.
When publishing your app, simply choose your desired access level and we will prevent unauthorized users from reaching your app.
http://Lightsprint.ai has created hosted GStack, so it's the fastest way to get going on creating new things
Lightsprint(YC P26): Lightsprint now supports Gstack! Go to https://lightsprint.ai/gstack to try out our design shotgun and implement the plan in one-shot
▲ ~/ 𝚗𝚙𝚡 𝚍𝚎𝚎𝚙𝚜𝚎𝚌
John Resig: deepsec is very cool! I ran it against our main backend repo at @khanacademy and it found some really interesting edge cases. Spun up an agent team in Claude and addressed all the high-impact issues. This likely would've been a multi-week, or month, effort otherwise!
Replit ⠕
Now all Replit builders can securely connect webhooks, GitHub, Slack and other external services to private apps
So an internal tool can post Slack updates, or a business app can receive Stripe payments, while staying private to your org or team.
To enable this, publish your app privately, and you'll see an External Access Tokens section under Security in the Publishing pane, where you can create and manage tokens for these external services
