Jake Taylor

For any quantum gravity aficionados at APS, we are having a session on experiments — tabletop and quantum simulation — this afternoon.

Sad news – just learned that Tony Leggett, visionary condensed-matter theorist (and in passing, Nobel laureate) has passed away.

Protesting or at least opposing data centers is highly rational for everyone except the most direct equity holders hence beneficiaries of the AI and AI value chain companies. AI has a political constituency problem. Most people know that they are net losers from the trade off of “losing your job/economic value” in exchange for maybe “cure for cancer”, a free robot, UBI and permanent underclass status. They have also been sold pretty convincingly, by the spiritual leaders of the AI industry no less, that this was inevitable no matter what they did. At most they could stall it and prevent it from happening right away. In your typical lobbyist group issues, the benefits are highly concentrated among a few highly motivated and the costs are spread thin across a majority that’s unmotivated and not paying attention. As such the net losers typically have trouble organizing as an effective opposition. AI though has everyone’s attention and has made most if not all acutely aware of the existential stakes. Ask a random person off the street the first thing they’ll tell you about AI is that it’ll take their jobs. I’m not sure we’ve ever had an issue like that. The funny thing about the “AI inevitability” is that it still requires astronomically if not comically high economic cost with significant negative externality to come into being. The average person in a democracy could exercise their political rights to delay the “inevitable” at least in exchange for a stake. It’s perfectly rational, incentive compatible and shockingly they do.

Now we are actually going to do it. newscenter.lbl.gov/2026/02/18/fiv…

I will be enthusiastically supporting faculty legislation to cap the number of A's at Harvard at 20% (plus a bit). The collective action problem that has driven grades higher & higher over time is increasingly problematic. I hope other institutions consider similar steps.

CAISI is hiring for a bunch of exciting new roles, from partnerships to technical experts in AI x bio / chem and more. They're serious about bringing in strong researchers & engineers and letting them do good work. Based in DC or SF: nist.gov/caisi/careers-…

Today we’re releasing the International AI Safety Report 2026: the most comprehensive evidence-based assessment of AI capabilities, emerging risks, and safety measures to date. 🧵 (1/17)

people don't appreciate 1e-18 fractional accuracy enough. It took decades of development pushing many things to the extreme. For example, let's take a look at the optical cavity used for stabilizing the laser used for these atomic clocks. The most recent fight they had with the cavity is replacing the dielectric coating (sputtered SiO2/Ta2O5 or TiO2), which is amorphous, to stacks of crystalline GaAs/AlGaAs, because crystals have lower thermal mechanical noise than amorphous materials, and they got a cavity with 2.5e-17 stability with such quieter mirror coatings. [Lee2026] How good is 2.5e-17? When you are here, you are at the extreme opposite of "nothing ever happens". Everything is happening, and everything affects you. Temperature? It gives you at least two big headaches, (temperature fluctuation) * (thermal expansion), and thermal noise itself. So you optimize the cavity shape, you use single crystal for both the mirror spacer as well as the mirrors themselves so they are less lossy and thus less noisy, and you bond them together along the same crystal orientation as closely as possible. You also cooldown the cavity to reduce thermal noise, not only that, but you also operate at the CTE zero crossing point so temperature fluctuation matters less, thats why you see 124 K and 17 K for silicon cavities. Even with zero CTE, you still need few mK temperature control. Away from zero CTE, it may need to be stabilized to sub uK level. If you glance into cavities working at 4 K (i.e. small but not at zero CTE), you'll see crazy thermal damping systems to smooth out the temperature fluctuation of the 4 K cryostat (~20 mK), as well as find claims like "we now require only mK level control of the room temperature enclosure". [Zhang2017, Robinson2019] (they said "only" because some older cavities were at room T and controlled to sub mK [Ludlow2007], and they got tricks to reduce effects from room T.) (temperature gradient also gives you headache, which is why they choose silicon over glass, for its much higher thermal conductivity) Next is vibration/acceleration. Nothing is rigid, the cryostat vibrates, the earth rotates, and your cavity changes shape. Thats another motivation for silicon over glass, for its higher Young's modulus. So you make the cavity shape as symmetric as possible, and make the mounting fixture as symmetric as possible, and align them with the crystal axis because silicon's Youngs modulus not isotropic . Thus you also choose the optical axis to be the crystal axis with the highest Young's modulus. [Kessler2012] You also gotta align the mechanical axis with the optical axis and with the crystal axis, otherwise longitudinal acceleration would tilt the mirrors and change cavity length. The spacer shape is also a double cone so that it sags less and bends less under transverse acceleration. [Millo2009] Any mechanical resonance would be bad, so you also gotta use PEEK instead of PTFE for supporting the mounting ring, and push the lowest mechanical resonance to be as high freq as possible. Remember silicon's crystal structure? Remember its 3-fold rotational symmetry? That's why your support structure also has the same 3-fold symmetry. [Matei2016] If you have done all these properly, congratulations, now you might be ready to start fighting thermal noise in the dielectric mirror coatings. Harry2002: [Thermal noise in interferometric gravitational wave detectors due to dielectric optical coatings](doi.org/10.1088/0264-9…) Numata2004: [Thermal-Noise Limit in the Frequency Stabilization of Lasers with Rigid Cavities](doi.org/10.1103/PhysRe…) Ludlow2007: [Compact, thermal-noise-limited optical cavity for diode laser stabilization at 1e-15](doi.org/10.1364/OL.32.…) Millo2009: [Ultrastable lasers based on vibration insensitive cavities](doi.org/10.1103/PhysRe…) Hopcroft2010: [What is the Young's Modulus of Silicon?](doi.org/10.1109/JMEMS.…) Kessler2012: [A sub-40-mHz-linewidth laser based on a silicon single-crystal optical cavity](doi.org/10.1038/nphoto…) Matei2016: [A second generation of low thermal noise cryogenic silicon resonators](doi.org/10.1088/1742-6…) Zhang2017: [Ultrastable Silicon Cavity in a Continuously Operating Closed-Cycle Cryostat at 4 K](doi.org/10.1103/PhysRe…) Robinson2019: [Crystalline optical cavity at 4 K with thermal-noise-limited instability and ultralow drift](doi.org/10.1364/OPTICA…) Lee2026: [Frequency Stability of 2.5×10^−17 from a Si Cavity with AlGaAs Crystalline Mirrors](doi.org/10.1103/zgrm-c…)

Looks like some consolidation in quantum industry

When we started Axiomatic_AI, we were convinced that a form of math-grounded AI would eventually ace the Physics Olympiad; that we'd get there within 18 months is still a huge surprise and amazing accomplishment by the team. Congratulations everyone ! linkedin.com/feed/update/ur…

When the team came back with this serendipitous result I was flabbergasted. We haven’t designed lemma for this, but it apparently comes naturally from the approach we take (see our AxProver paper for the general concept). linkedin.com/posts/axiomati…

Recommendation letters for the Hartree are due on Dec. 8th. Please verify with your writers that your letters are uploaded!

Terry Tao notices Gemini DeepResearch off-handedly solve Erdos problem #481 while doing a literature review. Gemini doesn't appreciate its own success and proceeds to pontificate on why the problem is hard.

We are in the beginning of a new era of scientific discovery powered by AI. Today @POTUS launched the Genesis Mission, uniting @ENERGY’s national labs, vast federal scientific datasets, & American researchers to accelerate science & tech breakthroughs. whitehouse.gov/fact-sheets/20…

FP64 is not dead. Do you hear me NVIDIA? Emulated FP64? Ozaki scheme? Please discuss!

Great thread that basically summarizes my thesis topic

Proud of the team @Axiomatic_AI — we have the best open source prover on PutnamBench, and third best overall, for pennies on the dollar when compared to Hilbert:

Delightful seeing John post-prize at AQC ‘25, joined by his once postdoc Pedram

I keep warning that so many of our systems are still built around the assumption that quality writing and analysis are costly and therefore meaningful signals. Our systems are very much not ready for the revelation that this is no longer true, as this planning objection AI shows

It is utterly fascinating to me the number of people making an argument for some type of de facto govt subsidy for the frontier labs by drawing comparisons to (for example) nuclear energy without being willing to accept ANY of the regulatory framework core to that same industry.