Max Curnin (@MaxCurnin) header
Max Curnin (@MaxCurnin) profile picture

Execution went electronic. Authority did not. backing the atoms those agents will trade exec chairman @Accordsai · 021 https://teapaw.org/t.co/UpAoukGCdm

Miami, FL
Joined July 2017
Photo by Max Curnin (@MaxCurnin)Photo by Max Curnin (@MaxCurnin)Photo by Max Curnin (@MaxCurnin)Photo by Max Curnin (@MaxCurnin)Photo by Max Curnin (@MaxCurnin)Photo by Max Curnin (@MaxCurnin)Photo by Max Curnin (@MaxCurnin)Photo by Max Curnin (@MaxCurnin)Photo by Max Curnin (@MaxCurnin)Photo by Max Curnin (@MaxCurnin)
Picture a trade five years out. Your agent rebalances. It hits an exchange's agent. That agent pings a liquidity provider. A market maker hedges the other side before your order clears. Four institutions. Zero humans in the loop. The thing being traded is increasingly compute itself. Every handoff crosses a wall. Whose rules apply. What it can commit you to. Who is accountable when it is wrong. That is not a security problem. It is a governance problem.
4
748
Max Curnin retweeted
Well, looks like @__alpoge__ & @claudeai are claiming a proof of the Riemann Hypothesis alpo.ge/remark.pdf
178
785
594
6,586
1,312,432
It's clear now more than ever: Ultimately, this bear market in tech stocks will be remembered as the best buying opportunity of the last decade. Unless your view is that the AI is worthless and the AI Revolution is over, many key parts of this market are extremely cheap. Democrats will be more friendly to AI than expected, Nvidia will survive its leadership crisis, and concerns around CapEx and free cash flow will prove to be overblown. To top it off, current sentiment is as if the market has just gone through another 2008, all while the S&P 500 is posting +32% earnings growth. The comments on this post will show you just how polarized sentiment has become in the direction "bubble" callers. Fear creates an investor's biggest opportunity. Bookmark this post.
1,023
1,495
269
18,756
1,966,520
🫡
Responding in my personal capacity, but wanted to share my perspective.. We are in a great-power competition with so much at stake in the years ahead. There is no one in America who has lived during a time when the United States wasn't the preeminent superpower and, despite our imperfections, a force for good in the world. A nation that fought and helped end wars on behalf of those who couldn't, pioneered powered flight, put humanity on the Moon, led revolutions in medicine and technology that improved billions of lives and distributed hundreds of billions in aid around the world. I don't know how handicapping this nation by going after one of its largest employers, a company that has created wealth for millions, ushered in electric cars, rockets that don't need to be thrown away, satellites that connect the disconnected and have unquestionably saved lives, not to mention access to information that is pretty foundational to solving many of the world's problems, biotech to restore function to those with disabilities, and so much more, serves America's interests. Without the contributions of these companies, there is little question America would be behind in a competition where the outcome is far from certain and the consequence so great. Now, maybe I am just a bit new to this, but from the billboards, TV and digital ads, spam texts, and fundraising calls, it would seem both parties have no issue raising money, including from billionaires, with the aim of influencing and buying votes. And if that means campaign reform, maybe include dialing back the spam calls and texts..But with all that is going on in the world, and so much at stake for our country, I question the priorities reflected in the post below.
39
Hyperbolic bought a put on B200 rent Tuesday: ~$5M notional, one year, settled monthly against Silicon Data's on-demand index. Same day, Nvidia's Colette Kress said neoclouds are growing faster than their balance sheets. I'd watch for a GPU loan with the put in the collateral.
48
Max Curnin retweeted
Bringing AI to mineral exploration is one of the most obvious and compelling applications I have seen, and yet very few teams are focused on it. This week may be a celebration of LLMs. but meanwhile @earthaiexplore is using their models to find literal gold.
In 1969, explorers drilled an Australian copper-gold deposit and wrote it off. 57 years later, @EarthAI has outlined a system stretching over 3km at Fontenoy, NSW Our new AI model, Alpheus, has flagged 4,500 more targets, giving us the potential to double the world’s mines 🧵
3
3
23
1,072
Max Curnin retweeted
In 1969, explorers drilled an Australian copper-gold deposit and wrote it off. 57 years later, @EarthAI has outlined a system stretching over 3km at Fontenoy, NSW Our new AI model, Alpheus, has flagged 4,500 more targets, giving us the potential to double the world’s mines 🧵
74
112
19
1,482
1,091,133
Max Curnin retweeted
I just love Earth AI. They just keep finding valuable metals and proving out deposits. Big list to go after. More money —> more rigs —> more discoveries —> more money —> more rigs, all the way down the target list. G’day for Earth.
In 1969, explorers drilled an Australian copper-gold deposit and wrote it off. 57 years later, @EarthAI has outlined a system stretching over 3km at Fontenoy, NSW Our new AI model, Alpheus, has flagged 4,500 more targets, giving us the potential to double the world’s mines 🧵
11
5
1
90
12,522
Max Curnin retweeted
Visit to @pilgrimlabs today. @jakeadler is going to change the world.
1
3
44
1,534
Max Curnin retweeted
We are so much more vulnerable than we think to bio weapon’s. Very bullish on @pilgrimlabs and @jakeadler. He has the makings of a generational founder.
2
3
1
26
3,801
Max Curnin retweeted
Great to have you visit 🫡 excited to show off more of ARGUS
We are so much more vulnerable than we think to bio weapon’s. Very bullish on @pilgrimlabs and @jakeadler. He has the makings of a generational founder.
6
4
56
2,592
Max Curnin retweeted
Neurophos CEO Patrick Bowen is speaking at TomorrowX Summit, joining a lineup including AMD’s Lisa Su, Intel’s Lip-Bu Tan, Arm’s Rene Haas, and Micron’s Sanjay Mehrotra. ⃗See the lineup and register: tomorrowxsummit.com November 17–18 at the Moody Center in Austin. @TomorrowXSummit, @PatrickTBowen
🤖 Made with AI
7
15
806
Max Curnin retweeted
Every project and field trip we do, unfolds many million years of history about our Earth 🗺️ At Earth AI, we are revolutionising exploration so that we can learn faster without compromising accuracy #exploration #geology #EarthAI #mineraldeposits #minerals
1
4
457
AI Demand Is Real: Prices Halved, Volume Exploded Skeptics say AI is a capex bubble with no paying customers. The usage data says otherwise. Weekly token volume on OpenRouter hit 210 trillion last week, up 8% in a single week. A year ago it was in the low single-digit trillions, so volume has grown by a factor of 40 or more in twelve months, and this week is pacing at 215 trillion. Meanwhile, the price of AI collapsed. Silicon Data’s LLM token expenditure index fell from about $2.00 to about $0.80 per million tokens over the past year, and Anthropic’s weighted price dropped by more than half. When something gets 60% cheaper and usage goes up 40-fold, total spending explodes. That’s Jevons in real time: cheaper compute creates more demand for compute, not less. The mix is getting heavier too. Memory-intensive long-context requests rose 11% in a week to 616 million. Those are agents, codebases and documents, the workloads that eat GPUs and memory. Proprietary models are holding near $2.00 per million while open models sit at about $0.50, so customers are paying for quality. That’s pricing power, not a race to zero. This is AI moving from expenditure to evidence. Falling unit costs and exploding volume is the productivity story, the same supply-side force driving unit labour costs lower. The capex is chasing demand that’s already here.
20
44
3
245
11,859
I get asked very often on something that I would call the optical accumulation problem. This was actually one of the reasons why many people abandoned optical compute in the 1980's (see opg.optica.org/ao/abstract.c…, it’s about DMAC obviously but the math and fundamentals are the same for matmuls), and why some others have deemed it impossible today. This problem is also something that, to the best of my knowledge, Neurophos solved for the first time in history. The optical accumulation problem is basically this. Say you’re running matrix multiplication on a 4-bit signed integer machine, so your numbers run from -8 to 7. Then to do the matmuls, you need to multiply two numbers together. That means your smallest number (in magnitude) is -56 and the largest is 64. Then, if your tensor core is let’s say 1k by 1k, the next step is that you need to add 1000 such numbers together. So now the smallest number is -56,000 and the largest is 64,000, which is a range of 120,000 numbers that you need to be able to resolve. If you want to sample an analog signal that should resolve 200,000 numbers then you run into two seemingly insurmountable problems. First, you need a 17-bit ADC to sample the signal. The problem is the energy for an ADC to sample a signal goes as 4^B, and if B is 17, that ends up being (see Murmann data) around 1.5nJ. A 3nm/4nm GPU's tensor core’s digital MAC costs around 0.2pJ, so even though that 1.5nJ represents 1000 MACs, after it’s amortized it costs 1.5pJ, and you’re nearly 10X LESS efficient than digital. The second problem is that you’d need at least 200,000 squared photons per accumulation due to shot noise, or 10^10 photons, which is around as much energy again per MAC as the ADC. So at first glance, from fundamental physics reasons, the minimum performance of optical compute would seem to be about 20X worse than digital. This is not only a killer for optical compute, it’s a death blow to any kind of analog compute, unless you can come up with a serious solution. I’ve met people who abandoned optical or analog compute for this reason, but I’ve never seen anyone else solve it. Neurophos fortunately did. We hired an incredibly bright mathematician from MIT, and I worked very closely with him to find the solution. Neurophos’ solution involves special pre- and post-processing of the digital data coming into the tensor core at very low overhead, and a custom number system to represent the numbers in the optical middle. The number system part of the solution is related to old information theory and number systems that we had to modify for optical compute. The result was that we were able to design an optical tensor core that not only gets around this problem, but delivers digitally equivalent precision and accuracy as a digital tensor core. After we solved it, we realized that fundamental math was always going to be an important part of any future compute product, so we wanted to invest in math as part of our company’s future. That is why we announced our collaboration with @MIT’s @MIT_CSAIL. csail.mit.edu/news/collabora…
3
5
14
581
Max Curnin retweeted
America’s biodefense infrastructure is long overdue for an upgrade. Spoke with @KR_Callaway and @nytimes about where the current system is failing and what comes next
2
10
1
46
3,272
Max Curnin retweeted
When did @TheEconomist staff writers become a mouthpiece for Chinese national interests?
China wants Jensen out, so I want him in.
54
158
14
1,799
224,520
Max Curnin retweeted
In case you missed it, DoE's safety analysis approval is a huge milestone for Radiant and nuclear energy. This is the first approval for a private portable microreactor. Kaleidos has no radioactive liquids, a minimal land footprint, and produces usable dry heat at ~1,200F.
Radiant has secured U.S. Department of Energy (@ENERGY) approval of its Documented Safety Analysis for Kaleidos, making it the only microreactor to be reviewed and approved by the DOE for full power operations. The approval is not just for criticality, but for Radiant’s full power, full temperature testing at the Idaho National Laboratory. Radiant will test the complete reactor system using the same design it will manufacture at scale. Next up in the path to commercialization: Readiness Review. Read more here 🔗 radiantnuclear.com/news/radi…
3
5
33
3,204
402Sentinel retired its hosted pre-payment check on Oct 31. The package can still make x402 payments, and v0.9.0 refuses 402sentinel.com. Check which version your agent pinned. 7 checkers compared, the retired one included. I run this directory. theagenteconomy.si/compare/a…
31
Max Curnin retweeted
Zero-shot long-horizon, whole body manipulation is here. Heard from many people that this is the most impressive demo they saw at CORL this year :) Come check it out!
Yesterday, we unboxed Taku at @corl_conf put it to work. Powered by Dyna-2.1, it performed long-horizon, whole-body mobile manipulation zero-shot, with no on-site data collection or post-training. Come see it live and meet the team at Booth 500 in Griffin Hall!
2
6
87
4,340
Anthropic mostly rents its compute. Now it's paying about $3.4B for 1.7 GW of gas generators, taking over Caterpillar orders OnePWR placed two years ago and couldn't finance. None of OnePWR's staff came with it, so running them falls to Anthropic. cnbc.com/2026/11/08/anthropi…
30