Michael Hendricks retweeted
Replying to @kimmonismus
Something for you - Anthropic employee Levent Alpöge has just claimed to have proofed the Riemann hypothesis with Claude: alpo.ge/remark.pdf
Michael Hendricks retweeted
We’ve priced our IPO at $83 per share, with trading expected to begin on Nasdaq tomorrow.
anthropic.com/news/anthropic…
Legal disclaimer: A registration statement relating to the Class A common stock was declared effective by the Securities and Exchange Commission on November 18, 2026. The offering is being made only by means of a prospectus. Copies of the final prospectus may be obtained, when available, from: Goldman Sachs & Co. LLC, Attention: Prospectus Department, 200 West Street, New York, NY 10282, by telephone at (866) 471-2526, or by email at [email protected]; or Morgan Stanley & Co. LLC, Attention: Prospectus Department, 180 Varick Street, 2nd Floor, New York, NY 10014, or by email at [email protected]
Michael Hendricks retweeted
Lot of people asked about this. You can basically make your MCP self improving by doing this:
- Add a submit_feedback tool to your MCP
- Agent using your MCP will write detailed issue including proposed fix.
- Have it send to an internal Slack channel
- Have Claude in Slack channel and tag it to put up a PR when you agree w proposed change.
This is why you should actively invest in an embedded agent and an MCP.
MCP observability only allows you to see tools called and you have to reverse engineer intent.
If you have an actively used embedded agent, you would have seen this query and been able to ID that the proper filters were not in place. Fix the tool and it benefits both surfaces.
You can also get a lot of benefits by adding a submit_feedback tool to your MCP to gather this data.
Michael Hendricks retweeted
Transformers have barely changed since the 1890s: copper coils in a tank of oil.
Batteries, solar and compute run on DC. The grid runs on AC. Bridging the two has meant inverters plus transformers.
Powercore is a solid-state transformer, breaker and inverter in one box. Finally, one conversion and about half the losses.
Hats off to the Tesla power electronics engineers!
Michael Hendricks retweeted
Our Cloverdale, CA Supercharger is a PG&E Flex Connect site, talking to the grid over IEEE 2030.5 to open faster.
Michael Hendricks retweeted
Announcing a major milestone achieved by my team @GoogleDeepMind: we have proven key security properties of the machine code underlying the Signal messaging protocol. This is the first time the actual production code of a widely deployed protocol has been formally verified – by any method, with or without AI – and at this depth.
More than a research breakthrough, it’s a prime example of how frontier AI can be a force multiplier to counter the speed and scale of modern cyber threats.
Formal verification offers a higher standard of security; however, it’s extremely complex and time consuming, which is why in many cases it’s considered to be impractical. Similar efforts could take teams decades of combined effort. We did it in a month’s time.
Using AlphaProof Code, Google DeepMind's new agentic framework for formal verification that runs on LLMs including Gemini Carbon, we verified the Signal protocol down to the actual code running on devices. This milestone lays the groundwork for a future in which vital systems, from medical device firmware to power grid software, ship with mathematical proof that their protections work.
Reaching that future will take continued research from our community. In that spirit, we're open-sourcing our formal specifications and verification artifacts so others can build on this work.
On a personal note, this moment is of deep significance to me. I started my own research journey more than two decades ago as an intern working on checking the consistency of complex software systems. Seeing this grand challenge made practical with modern AI is extraordinary.
I'm extremely proud of my team members: Swarat Chaudhuri, Andrew Ferraiuolo, Anton Kovsharov, Tej Chajed, Sam Schlesinger, Karn Seth, Jonathan Katz, and Charles Sutton for taking on this grand challenge and delivering, and grateful to the Signal team for their partnership.
deepmind.google/blog/formal-…
Michael Hendricks retweeted
i've reproduced the results from the MIT blog post today on weight-space programming. but one additional result i have to add: this is a new, step-change emerging capability that's appeared in just the latest set of models.
Michael Hendricks retweeted
Just finished up, here is the real Houston Geofence that is ~26% smaller than advertised.
Waymo is next, and they're no better.
Replying to @ethanmckanna
It’s a decently large area.
Think it’s time to dig into every service area and find out what’s real
cc @koby_pierce
Michael Hendricks retweeted
3) Imagine that tomorrow we invent a $10k robot that can build brick walls. What does it do to growth? Very little! Doesn't affect brick prices (can even raise them in SR!) or "roll fee" for the truck that brings it to you. But it's worse. 2/8
Michael Hendricks retweeted
I don't care if I sound like a shill, my Tesla is amazing
Every drive is just, wow, the car is driving
Soon drunk driving manslaughter will be over. Text-and-drivers, geriatric loons on the wrong side of the road, road ragers, just gone, a historical problem
Michael Hendricks retweeted
🚨 WE HAVE A PERMITTING DEAL🚨
And it is easily the best permitting bill we've ever gotten!
- The Bipartisan American Affordability and Jobs Act is HUGE (400+ pages) and both sides got a LOT.
- The bill is much bigger and better than either Manchin post-IRA in 2022 or Manchin-Barrasso in 2024.
Once I finish reading the text I'll publish an explainer. For now, here are a few high level takeaways.
1. Transmission: The bill is absolutely HUGE for transmission, going further than Manchin-Barrasso. Includes: interregional transmission, limited federal backstop authority, interconnection queue reforms, directives for advanced grid-enhancing tech, and oversight of utilities. It's absolutely AMAZING all of this got in.
2. NEPA: The bill has BIG changes to NEPA, including fixes for judicial review that underpins the law's costs. Still reading but this is even better than I expected!
3. Certainty: The bill includes strong provisions to push back on misuse of permitting, like we're seeing with the admin's abuses against offshore wind.
4. Historic Preservation: The bill includes changes to the NHPA 106 process, including changes to judicial review that will help right-size the process.
5. Clean Water: The bill fixes problems with section 401 by adding new requirements for states and makes section 404 nationwide permits easier to comply with.
6. Geothermal: The bill includes big wins for geothermal, including parity with oil and gas catexs.
7. Mining: The bill fixes the "Rosemont decision"
8. Endangered Species: The bill creates a new path for states to take on endangered species consultation.
9. Hydro: Fixes for certain hydroelectric projects
10. Ratepayer protection: The bill adds protections for ratepayers from added data centers. Looks to be stronger than the House passed bill by quite a bit.
Michael Hendricks retweeted
Claude Code will now try to find a graceful stopping point when you hit your 5-hour limit mid-task, instead of cutting off mid-edit. It gets a small, fixed allowance pulled from your weekly limit to wrap up what it can.
Michael Hendricks retweeted
New Tesla Supercharger: Lander, WY (8 stalls)
tesla.com/findus?location=19…
Michael Hendricks retweeted
David Deutsch (@DavidDeutschOxf) has a nice post today explaining how he fixed his dishwasher with the help of ChatGPT. We have all been there over the last couple of years. In my case, from fixing a light fixture in my kitchen to repairing a leak in my irrigation system.
Deutsch and many readers point out something that has been a staple of my talks for a decade. GDP was designed to measure an economy that produced coal and steel, easily measurable quantities. And GDP did its job quite well for decades.
IT, first, and later the internet and AI, changed the landscape. My favorite example is books. Even as late as 2010 or so, getting the most recent books in Spanish on, let’s say, Spanish economic history was hard in the U.S.: expensive and time-consuming. Every time I went to Madrid, I carried two bags, one of them empty because it would come back filled with books. Then Kindle arrived, and now I can buy nearly every single Spanish book I want in seconds and at a lower price than the physical copy. This was an enormous increase in my welfare that never got reflected in GDP.
The literature is divided on how big this effect is. I tend to side with those who think it is a big deal. Therefore, many of the gains in GDP growth announced by the strongest defenders of AI might not materialize: they will go instead into higher welfare.
But the issue goes beyond getting GDP wrong. What Deutsch perhaps did not notice is that he did not pay VAT on the service of the repair person he did not hire, and that such a repair person did not pay income taxes on the service he did not render.
Taxing AI-based economies might thus become much harder, from a purely technical perspective. One solution would be to tax the profits of OpenAI (or Deutsch’s subscription), but I am among those who think those profits are not going to be nearly as large as anticipated. Competition among labs will drive the subscription prices of models down (unless labs can cartelize, for example, by appealing to “AI security”). Over 95%, and perhaps as much as 99%, of the gains from AI will pass to the final user, whether another business or an individual.
Now, we could tax the final user, for instance, at a fixed rate per token used. But that implies we are taxing the high-return activity (how to fix the dishwasher) at the same rate as the low-return activity (Claude: “tell me a joke about my cat jumping”). And taxing per token might hurt those final users who use more tokens because they are more productive, something we might not want to do (yes, there is something called the uniform commodity taxation theorem, but it often breaks down).
Let me give you an extreme example: imagine that tomorrow the new OpenAI model cures all cancers, forever. Just a vaccine, $10, once in your life, and you never get cancer again. The increase in human welfare from such an invention would be staggering. But how do you measure it in GDP? Yes, people will live longer, perhaps work longer. A second-order effect on GDP. And more importantly: how do you tax the vaccine? Do you even want to tax the vaccine? I would argue we want to subsidize it, and generously.
When I was still an undergrad, my first RA job was helping to construct GDP and other national aggregate statistics in Spain. It was a most valuable experience, because it made me aware of what GDP means, its strengths (and it has many more than its critics concede), and its limitations. But it also taught me that what you can tax out of GDP is far more limited than you would think at first sight.
There are real upper bounds on taxation, and I am afraid the bound is tighter now than it used to be.
I just fixed my dishwasher with the help of ChatGPT. A trivial task. I had been about to order a new one. So this software has increased the country's real wealth yet decreased the measured GDP. The main economic indicator is structurally incapable of registering the thing that actually makes people better off, namely the growth of knowledge





























