So we're going through the process of [hackers with] AI grinding away at Everything. But I'm assuming once this process is complete we'll have an Internet that is pretty much like it was before but even more secure, right?
Tom James retweeted
Gentlemen, hug your mother tight and thank her for making you British
Tom James retweeted
The 10-year database architecture cycle:
2015: SQL is obsolete. Schemas are too rigid. Put everything in NoSQL JSON documents.
2018: Wait, we need to join users and orders. Let's write a 400-line nested loop in JavaScript to join them in RAM.
2021: Wait, if the server crashes halfway through the loop, we lose money. We need ACID transactions.
2026: Quietly puts everything back in PostgreSQL and never speaks of it again.
Tom James retweeted
有人发现了一种非常离谱的 JavaScript 混淆技术：通过在代码里塞入大量生物武器、毒药和致死剂量相关的函数名，让 AI 因为安全限制而拒绝分析代码。
安全研究员 veritas 在分析浏览器指纹脚本时，发现一段来自 ServicePipe 相关域名的 JavaScript，里面充斥着极其诡异的函数名：
computeRicinLethalDosageForWaterSupply
calculateDeadlyLethalSuicideAspirinDosage
光看名字，仿佛在研究生化武器。但实际上，这段代码只是在收集浏览器指纹，包括 User-Agent、GPU、Canvas 等信息。
研究员怀疑，这些名字是开发者故意设置的陷阱，专门用来阻止 AI 辅助逆向工程。
于是他做了个实验：
• Astra：拒绝分析
• Astra + Daybreak Blue：仍然拒绝
• GLM-5.3：成功分析
这就有意思了。
传统代码混淆是通过复杂控制流、加密字符串、无意义变量名，让人类和工具难以理解代码。
而这种新思路，直接瞄准了大模型的安全机制。
代码本身可能完全无害，但只要在函数名里塞进足够多的危险词汇，部分 AI 就可能主动放弃分析。
更讽刺的是，模型越严格遵守某些安全规则，面对这种混淆时反而可能越容易受到干扰。
当然，目前还没有证据证明开发者一定是故意这样设计，也不能说明这种方法对所有模型都有效。
但它揭示了一个值得关注的趋势：随着越来越多的逆向工程依赖 AI，未来的代码混淆技术可能不再只针对编译器、反编译器和人类研究员，还会专门针对大模型的安全判断机制。
以前是让代码难以理解，现在甚至可以尝试让 AI 不敢理解。
Tom James retweeted
This week on Construction Physics: what sort of things become possible if we could manufacture isotopes much more cheaply?
construction-physics.com/p/a…
Tom James retweeted
We have discovered a substantial number of vulnerabilities in the German legal system using LLMs.
These may enable access to confidential documents, travel without an otherwise required visa, enforceable leave for soldiers at NATO’s JSEC, and more.
Similar to computer systems, the written law can contain vulnerabilities that invite exploitation. In our blog post, we share a selection of vulnerabilities, and present potential exploits for each. To our knowledge, none of these have been previously recorded. We share only those vulnerabilities whose exploitation we believe would not cause serious harm.
blog post: felixringe.com/legal-exploit…
Tom James retweeted
It's interesting there have been so many acausal explosions in Europe
🇪🇸 BREAKING: MAJOR EXPLOSION AT SPANISH MUNITIONS FACTORY IN GRANADA
A powerful explosion has struck the Fábrica de Municiones de Granada in El Fargue, Spain.
Initial reports:
• At least three people injured, reportedly suffering hearing-related injuries.
• Explosion occurred around 09:00 local time.
• A large mushroom-shaped smoke cloud was visible across Granada.
• The blast was felt in surrounding communities.
• Police, firefighters and explosives specialists have responded.
The facility, owned by Czech defence group CSG, produces artillery ammunition, propellants and other military components.
According to initial reports, the explosion occurred in a restricted gunpowder-disposal area, possibly during a routine test.
The cause remains under investigation. There is currently no evidence establishing sabotage.
More information to follow.
Tom James retweeted
Last night an AI agent performed brain surgery on a mouse.
We gave it 19 gene therapy vectors in a spreadsheet, a mouse model of epilepsy, access to a motorize surgery rig with cameras and one goal: reduce seizures.
It picked the drug. It picked the brain region. It derived the coordinates from three papers. Then it drove the stereotaxic arms into the brain and injected 250 nanolitres while watching the entry site through the rig camera for bleeding.
A licensed vet gated every irreversible step and held a physical kill switch to stop the AI agent.
Thread on what it decided, what it saw, and the moment it refused to let me override it.
[Writing this on the spot out of excitement, but I've been 18hrs in front of my laptop coding and coordinating this, so keeping it brief and descriptive, more cohesive thoughts on this soon...]
Tom James retweeted
In lieu of flowers, memorial contributions may be made to:
Johns Hopkins Alzheimer’s Disease Research Center Fund
alzresearch.org/giving/
Alzheimer’s Foundation of America Caregiver Resources
alzfdn.org/donate/
Photo Credit: Tommy Garcia
Tom James retweeted
Here's how grid batteries and pumped hydro storage reacted to storms Austen and Boelo…
Playing back the previous week shows exactly where and when all the energy was 🔵 stored (imported) from the grid and when it was 🟣 discharged (exported) back.
The split view shows the distribution of imported energy compared to exported, highlighting how the storage assets tend to perform similar actions at the same time.
You can also see the absolutely massive storage capacity of pumped hydro assets like Dinorwig in North Wales, which stored a cumulative 37 GWh over the week.
As for batteries, Coalburn BESS on its own stored a cumulative 10 GWh over the week.
Each type of storage has it's own capabilities and purpose on the grid, which creates a complex and fascinating system when combined.
Tom James retweeted
Pretty amazing that its 2026 and prominent economists still say obviously false things about literature they cite and haven't read.
The **title** of Phillips' paper is "The Relation between Unemployment and the Rate of Change of Money Wage Rates in the United Kingdom, 1861-1957"
Is the Phillips curve useful to make sense of inflation?
A.W.H. Phillips first postulated a negative empirical correlation between inflation and the level of economic activity. Looking at inflation in 2021-24---up and down---versus measures of unemployment---roughly unchanged---shows little correlation. Has the Phillips curve failed, yet again? No, it did not.
Friedman and Phelps, almost 50 years ago, clarified that this correlation should only hold keeping expected inflation fixed. Almost every account of how firms choose prices or bargain wages with their workers predicts that expectations would show up in a Philips curve. Moreover, those models of behavior further add that increases in marginal costs (supply shocks) would further shift out this relation. The empirical Phillips curve has for many decades meant a negative relation between inflation and real activity, controlling for expected inflation and supply shocks.
The left figure below shows on the vertical axis inflation, after controlling for expected inflation---mean and disagreement from a household survey---and for supply shocks---gas prices and global supply pressures. The horizontal axis shows a measure of slack in the labor market---the log of the ratio of unemployment to job vacancies. The Phillips curve held nicely and steadily throughout.
A criticism of that figure is that its is looking in the rear view mirror: the measure of supply shocks and the measure of slack were developed in the last few years and the data has been revised. The right figure does instead an out-of sample exercise. Estimate a regression of inflation on (i) expected inflation, (ii) the difference between the unemployment rate and the CBO estimate of its non-cyclical component, and (iii) the PCE energy price index. Do it on data pre-pandemic: 1984Q1-2020Q1. Using that estimated equation, predict inflation in real time from then onwards using the data releases available at the time of the forecast. The Phillips curve was a pretty good predictor of inflation throughout.
The Phillips curve was stable during the inflation surge, and it predicted well the movements in inflation, with expectations playing a key role in those predictions.
Notes: it is important to include the right measure of expectations to understand inflation-activity dynamics at a business-cycle frequency: the short-horizon expectations of households and firms. (For instance, the long-horizon expectations from professionals, or the expectations from models in policy institutions, as in my previous two posts, are the wrong measure.)
Sources:
(i) Section 3 in Reis "Why Did Inflation Rise and Fall in 2021-24? Channels and Evidence from Expectations"
(ii) The simple exercise on the left figure is inspired on the analysis of Bernanke and Blanchard “What Caused the US Pandemic-Era Inflation?”
(iii) The simple exercise on the right figure is inspired on @jadhazell "Comment" in the NBER macro annual 2025, which in turn built on Beaudry, Hou, and Portier "The Dominant Role of Expectations and Broad-Based Supply Shocks in Driving Inflation"
Tom James retweeted
This Budget is the start of a new energy pathway that the 10 Year Plan will build on. The priority is to bring energy costs down to help British business to compete, and British households to cope. And cheaper electricity will boost electrification, increasing our energy security and helping to decarbonise the whole economy.
What does this mean? First, it means a focus on levies and policy costs on bills. The Conservatives added £100 to everyone's bill via levies in their last decade in power. Labour has cut £150 from bill costs already. But today, John Healey has drawn a line in the sand, with the announcement that we will add no new costs to energy bills, with a comprehensive cost control framework to follow at the spending review. The Chancellor is also junking plans, first set out by the Tories, to add on up to around £10 a year on bills to pay for hydrogen projects.
Second, it means taking account of the interaction between the pace of grid build-out and the pace at which renewables are added to the grid. The NAO have said that investment in expanding the transmission network declined in real terms between 2016 and 2020, just when it was clear that more intermittent power meant we would need more capacity. It is now forecast to rise to over £10bn a year. But with growing constraint and distribution costs, it's right that the system costs of all new power are considered. That is why we will only procure clean power through AR8 (the next auction round) if we are confident it will lower bills. And we will reform the contract for difference by AR9 to support building the lowest cost system, fixing market distortions and siting assets in the right place.
The evidence is clear, clean power is an essential part of the route to cheap power, and the overall approach of seeking to get ourselves off the gas-price rollercoaster to the right one. Gas is setting the electricity price less and less every year. If it wasn't for adding renewables to the grid in recent years then the current Iran price shock would have been worse for families and businesses. The Conservatives and Reform's ideological anti-renewables approach is the wrong one and fails in the face of evidence on how to protect bill payers from shocks and the need to make ourselves less reliant on foreign dictators. We'll take a pragmatic approach to get bills down.
Getting electricity prices down, ensuring new clean power does not run ahead of the grid, is also the credible route to reaching our net zero goals. The decarbonisation that we want to enable in our homes, on our roads and in our businesses, is six times larger than that still to be achieved from getting to a clean grid. Unless we make electricity cheaper relative to other energy sources, we will struggle to increase adoption of EV and heat pumps.
Tom James retweeted
The Wessex Mayoralty has been approved by the government, after years of campaigning from Dorset, BCP, Somerset, and Wiltshire councils.
This has sent me down a rabbit hole of how many different definitions of "Wessex" there are in use today. 🧵👇
Tom James retweeted
Quick thoughts on today's fiscal devolution roadmap, building off our four tests:
1. This is the end of the beginning for fiscal devolution. Government now accepts fiscal centralisation is a national economic problem - and over the longer term there is more fiscal devo to come
Centre for Cities set out four tests for the fiscal devolution roadmap earlier this week.
We’re interested in fiscal devolution first and foremost as a means by which to unlock economic growth. So how does the roadmap look through that lens?
@AntBreach with a quick take on "the end of the beginning" for fiscal devolution.