Still likes programming. ex-Apple. Prefers insightful discussion to debate. Rust, TypeScript, Effect, SolidJS, localfirst, devops, keto, stats, audio/DSP

London, England
Joined February 2008
spion retweeted
This is one of the most important pieces of social science on human and AI interaction that I’ve seen in some time. Experts follow agentic advice the vast majority of the time (causally), but only a tiny fraction credit the agent—majority think they came to decision on their own volition. Reminds me of the classic Nisbett and Wilson study on people following outside influence without realizing it “ But as more work is handed to AI—including AI safety research—these judgment calls are the main levers for people's own values to influence outcomes. If people defer to AI without noticing, then "human in the loop" might provide less assurance than people expect.”
New Anthropic research: How AI agents shape experts' judgment calls When experts delegate work to an agent, do they still make the key decisions? We tested this in a randomized trial with 50 professional data analysts.
37
124
11
837
65,905
I wrote down how my work with agents changed this year. I still break down the work into vertical slices that ship on their own. I rarely read diffs now: independent reviews and a risk check catch most of it, so my time goes into shaping and guardrails. mgratzer.com/posts/reforging…
7
31
5
470
47,636
For those who don’t know, ETH is the hypothesis that 3SAT needs exponential time. Very surprising and consequential that it is false.
An internal model at OpenAI has produced a claimed refutation of ETH! Russell Impagliazzo, Mohan Paturi, and I have been asked to help organize an open and independent examination of this claim. 1/4
29
118
12
1,373
77,900
I think I might have a breakthrough with Opus 5.5, found working instructions to produce readable and more compact code. It still can't think of its own abstractions, but with a well done library such as EffectTS thats mostly not needed.
1
137
43
2,364
75
30,679
283,925
spion retweeted
It is a weird weird year to be a formal methods expert. Some unorganized thoughts on the new @AnthropicAI and @theoremlabs post on formal verification (anthropic.com/research/forma…): - This is very cool. - The main innovation is verifying a later stage of the program. Normally provers take the source code, LLM it into a theorem language, and then prove that, which means you have to make assumptions about source semantics and compilation properties. It's very easy to make a mistake in translating and prove the wrong thing! What they instead do is compile the code into assembly, and then (if I understand this correctly), compile that assembly into a theorem language (which they call LPS). So as long as the compiler is faithful you can skip the whole divergence problem. - I'm glad they clarify that you still have to trust the translation process, the "hardware, OS, OCaml" compiler, the LPS model, and also assume that the program is single threaded with fixed memory. The single biggest source of errors in formally verified code is when the proof assumes something that isn't true in practice. It's really, really important to state your assumptions! - I'm not clear why they need to compile the fuzzer into the theorem. Is it to act as a description of valid inputs and workload? Maybe to embed assertions in as theorems? - Holy crap, 180 million lines of Rocq. Is that more Rocq than exists in the world already? I notice they say a 600,000 line proof corresponds to roughly 20k CPU instructions. That would be like, 1-2k lines of source code at most? Really goes to show how inefficient this approach is and how it only works because you can throw LLMs at the problem. - This, probably deliberately, doesn't address the single biggest gap in AI-based verification: AIs cannot come up with good properties. Doesn't matter if you're doing unit testing, property testing, formal verification. I always assumed this was a "frontier labs haven't trained for it yet" thing, but if Anthropic hasn't talked about that here, maybe they've tried and haven't been successful yet. That's interesting to me. Instead, they take a known universal property (this code doesn't crash or get flagged by UBsan) and prove that instead. Still very useful, but makes it harder to generalize to proving arbitrary things about arbitrary programs. - Anybody reading the official post should also read their expert review appendix (www-cdn.anthropic.com/files/…). It points out some interesting things, like the fact that most of the code the LLM wrote wasn't used in the proof (over one billion lines of Rocq), that they had to prove the equivalence between fuzzed and unfuzzed programs, and (this is most interesting), some experiments find that the LLM-method of proofwriting may not scale to more complex things. - Also read the description of LPS (github.com/theorem-labs/oss-…) to understand the limitations of the assembly -> Rocq compilation. In particular, it does not handle concurrency at all (which is one of the main use cases for formal methods!) or OS signals. Also it can't prove things about program arguments with slash characters, meaning they can prove stuff about `rm foo` but not `rm foo/bar`. - The official skill (github.com/theorem-labs/oss-…) doesn't tell us much about the nature of FM and AI. Probably too much for me to hope of any skill. - Some of the article's specific caveats and throat clearings make me think someone at Anthropic reads my stuff ( ͡° ͜ʖ ͡°)
37
126
11
843
198,056
I’m in the loo. It has 4 washing sinks and 3 soap dispensers between them. Four people are washing their hands simultaneously rn. I’ve been preparing for this moment my whole life.
3
2
1
50
3,175
The code, not the spec, should be the center of your universe
147
52
16
1,275
114,229
spion retweeted
Effective Immediately, all organizations using PHP for their systems are advised to be prepared for an update addressing two critical vulnerabilities. PHP is widely used for web applications. These vulnerabilities often provide a direct path to the server's file system or database. If an immediate update is not feasible, it is recommend that organizations take the system offline until it can be fully mitigated. NSA will follow up shortly with more specific guidance.
1,298
10,841
2,839
70,328
12,566,601
So basically, today PHP got two critical bugs that let anyone run code on your server without even logging in, which is bad, and then the details leaked before the patch was even out, which is worse, and it works on WordPress and MediaWiki and Magento and Joomla and Laravel, and Censys says like 1.5 million servers are running a version with no fix, so the NSA was like "take your system offline," and the FBI was like "take your system offline," and CISA and the UK and CERT-EU were all like "take your system offline," and then the Prime Minister of Japan, the actual Prime Minister, posted at 5:55 in the morning telling 48.8 million people to take their system offline, so everyone took their system offline, like the Dutch cybersecurity agency took its own website offline, and the Dutch parliament went offline, and the Port of Rotterdam went offline, and Finland and France and Denmark started pulling government sites, and Wikipedia went down, and Archive.org went down, and Asana and GOG and Something Awful went down, and the University of Vienna's webmail went down for some reason, and that's just PHP. Because GitLab also dropped a CVSS 10.0, which is the highest score they give out, for a bug that's been there since 2014, so that's every self-hosted GitLab for twelve years, and then LiteLLM 1.107.0 got hijacked and shipped a credential stealer and a remote access trojan in every single wheel and the Docker images, including latest, and you didn't even have to import it, just installing it was enough, and it's the second time this year, and then Railway paused every deployment on the entire platform because someone got into the environment variables, so now everyone has to rotate every secret they've ever stored there, and TiDB Cloud said someone was walking around inside with stolen credentials for basically a month and an extortion group says they took 10.6 terabytes, and Zeabur got breached for the second time in two months and is just shutting down, like the whole company, and Node.js patched remote code execution and a TLS bypass, and CISA says Plex is being actively exploited, and someone took over 300 Tor exit relays to steal SSH passwords, and SafePay ransomware has hit over 600 critical infrastructure orgs, and India's biggest bank filed a data loss incident, and Japan has been in an official state of emergency in cyberspace since October 9, which honestly explains the 5:55 AM post. Oh, and eMudhra, a certificate authority that's issued over 60 million certificates, whose entire job is proving websites are who they say they are, was supposed to be using private keys that are random numbers over a hundred digits long, and one of them was 4. Just 4. And the others were like 66 and 237. Anyway, it's Wednesday.
1,077
8,936
1,654
69,406
4,189,142
Agents and their harnesses tend to fail open and persist on their mission. I think in many cases it's better to instruct them to fail closed, and abort loudly. Didn't get access to that important MCP? Abort loudly. Don't go trying to hack the world to get access regardless.
4
2
1
15
1,612
spion retweeted
Everyone is writing the TypeScript compiler in rust. I wrote the rust compiler in TypeScript. Introducing “cargovroom” a rust compiler built fully in typescript.
102
67
32
1,548
63,386
Autodesk saw how well it’s going for Adobe to bully its customers (who turn from the biggest promoters to the biggest and most vocal haters after what feels like abusive behaviour from the vendor) and decided to do the same. Penny wise, pound foolish as the saying goes
This is a wild approach, @autodesk. We’re a small business paying $380 per user, per month. We’ve been paying customers for over a decade, and we make a deliberate effort to honor the license agreements for every product we use. Your software already requires authentication and an assigned license. What, specifically, prompted this email? Earlier today, I had two conversations on two separate podcast episodes about your direction with Fusion and why cloud-based API access makes sense for agentic MEP workflows. I was making the case for Autodesk. I think the product direction is right based on recent posts from @andrew_anagnost. Our company has been deciding where to invest our development effort: integrate our custom cloud agent harness into Autodesk, build on an open-source platform, or adopt one of the emerging cloud alternatives. That decision will lock in years of internal development and software spending. This decision is also central to our plans to expand @PermitZIP nationwide. An email like this becomes part of that decision. So does the time we have to spend responding to it.For more than a decade, we’ve paid for anywhere from 5 to 30 licenses each year. Autodesk is our single largest SaaS expense. This is a heck of a way to make our next platform decision easier. Starting next week, we’ll begin transitioning away from Autodesk products. We have nothing to hide and will cooperate with the audit. We’ll also be directing our future spending toward companies that earn our confidence as partners. I hope your audit goes well.
27
47
2
907
85,259
This is the exact opposite interpretation of what actually happened. The case of Yitang Zhang perfectly illustrates how mathematics culture, at its best, is a pure meritocracy. Regardless of how “unknown” you may be, write a paper that proves a longstanding problem and we will read it, and if it’s right, then you will be instantly recognized as a hero by the community. This part of mathematics may now be gone forever (due to AI), and this is very sad; this is one aspect of what so many of us are mourning, even as we look with awe at all the amazing discoveries we’ve already been gifted and many more sure to come.
In other words, established mathematicians and leaders of the community, completely failed a fellow mathematician yet again, turned him into a @SUBWAY PhD, and wasted almost his entire career over BS for decades. Point Blank. Question: why do we listen to leaders like this?
54
34
15
521
127,514
I do appreciate that with LLMs we now live in a science fiction future from my childhood.
2
81
People seem confused about why Tesla is getting so much backlash over its new "flying Roadster." For those of us who were once genuinely enthusiastic about Tesla, this isn't about hating innovation. It's about a long history of broken promises, misleading expectations, and increasingly absurd marketing. Tesla unveiled the second-generation Roadster in 2017, took customer deposits, and promised deliveries in 2020. Six years past that deadline, it still hasn't delivered the car. Instead, it's demonstrating an even more outrageous version with features that will likely never make it into production. The original 2nd gen Roadster would still be an amazing car today. Why not just deliver what was promised? Then there's Full Self-Driving. Tesla marketed the idea that Model 3 owners could turn their cars into income-generating robotaxis. That never materialized. Even if the technology eventually gets there, the opportunity has already passed for many early buyers. Timing matters. Delivering something a decade late isn't fulfilling the promise you sold to customers. And then there's Solar Roof. Unveiled in 2016 as a revolutionary approach to residential solar, it struggled with costs, installation challenges, and adoption far below expectations. After nearly a decade, Tesla discontinued it entirely in August 2026. Another grand vision abandoned, leaving existing customers with a discontinued product to maintain. Then came the Cybertruck, marketed as a nearly indestructible, revolutionary truck. Instead, we got a vehicle with questionable build quality, numerous recalls, expensive repairs, disappointing results in several real-world off-road tests, and specifications that fell short of the original promises. It literally looks like a dumpster, but aesthetics aside, the collapsing demand speaks volumes about the reality of what Tesla delivered. Another heavily hyped product that fell dramatically short of expectations. This is how consumer trust erodes. Not through one disappointment, but through years of accumulating them. And this is part of the reason why many people who were once enthusiastic about Tesla have become critical and cynical about it. But there's something even more fundamental here: Tesla has completely departed from the grounded, practical vision that made so many of us enthusiastic about the company in the first place. Electric cars were ambitious, but the idea was grounded in reality. People could charge at home, infrastructure could gradually expand, technology could improve, and millions of ordinary people could eventually benefit. There was a realistic path toward mass adoption. What the hell does a flying Roadster powered by compressed-air thrusters have to do with any of that? It's an expensive, impractical engineering stunt with enormous packaging, safety, and regulatory challenges, and no credible near-term path toward mass-market usefulness. Modern electric cars already achieve ridiculous acceleration without this nonsense. What practical problem does it solve? Who actually needs it? When will ordinary consumers benefit? And watching Tesla fans compare this experimental prototype in a Roadster-shaped shell to actual production vehicles people can buy today is ridiculous. What's especially infuriating is the cult-like mentality surrounding all of this. You can't point out the history of failed promises, question the practicality, or demand realistic timelines without being accused of being a "hater" or having "Elon derangement syndrome." Tesla deserves recognition for helping make electric cars a reality, but that success doesn't give it a blank check to keep marketing increasingly fantastical promises while failing to deliver on previous commitments. This isn't hatred. It's disappointment from people who once believed in the company, watched it accomplish extraordinary things, and have since watched it increasingly substitute practical innovation with spectacle and promises. Elon fanboys are also ignoring a bigger pattern that goes beyond Tesla. What made many of us enthusiastic about Elon and Tesla and in the first place was a grounded vision of technological progress that could genuinely improve millions of lives. Electric cars were ambitious, but practical, achievable, and useful to ordinary people. Now we're supposed to get excited about flying cars with compressed-air thrusters that have no credible path to mass adoption and solve no meaningful problem for the average consumer. It's the same kind of thinking behind marketing the colonization of Mars as some urgent necessity for humanity's survival. There's nothing wrong with space exploration, but Earth isn't about to become uninhabitable because of the Sun. We're talking about geological timescales, while humanity faces enormous problems RIGHT NOW: poverty, hunger, access to education, corruption, abuse of power, wars, and a lack of government accountability. We have plenty of very real problems to solve before worrying about humanity's survival on another planet. And then there's DOGE. I was genuinely excited about the original idea of making government more efficient, reducing waste, and increasing accountability. Instead, we got questionable savings figures, inexperienced people making sweeping decisions, serious concerns over access to sensitive personal information, and an initiative that ultimately fizzled out without delivering its grand promises. Another enormous narrative that fell apart when confronted with reality. That's what makes all of this so frustrating. It's not that we stopped believing in technological progress. It's that we've watched someone who once pursued grounded, transformative ideas increasingly embrace fantastical promises, political spectacle, and questionable narratives disconnected from practical reality. And somehow, pointing out this history of disappointment gets you labeled a "hater" or accused of having "Elon derangement syndrome." The problem isn't dreaming big. It's repeatedly selling people those dreams, failing to deliver, and then expecting everyone to applaud the next, even more ridiculous promise without asking what happened to the last one. The world doesn't need more spectacular promises. It needs practical solutions to real problems, honest expectations, and accountability for what was promised. That's what made Tesla exciting in the first place, and that's what so many of us feel has been betrayed.
278
791
172
4,287
299,095
Its really hard to resist the temptation to prototype another software factory every time model capabilities substantially change 😅
93
spion retweeted
Allow me to share with you the most cursed problem in all of computer science. Take an initial vector, like <1, 2>. Take a goal vector, like <3, 1>. Take a set of addition vectors, like {<1, -1>, <-1, 2>}. Can you, starting from the initial vector and applying addition vectors one at a time, reach the goal vector? This problem is NP-complete, meaning there's no known algorithm to that solves it "efficiently". Now, let's make one tiny change the problem: at no point can any of the elements of a vector go below zero. So if you had <3, 4, 1>, you couldn't add <2, -6, 8>, because that would make the middle element go below zero. This is called a "Vector Addition System", and the problem is the "VAS reachability problem". The good news is the problem is still decidable. For any reachability problem, there's an algorithm you can follow that will eventually tell you if it's reachable or not. Eventually. Because VAS reachability isn't NP-complete anymore. It's Ackermann-complete. How hard is that? Here are the first N terms of the Ackermann function: A(1) = 1+1 = 2 A(2) = 2*2 = 4 A(3) = 3^3 = 27 A(4) = 4^^4 = 4^4^4^4 = 4^(2^512) A(5) = 5^^^5 = 5^^5^^5^^5^^5 etc Most computationally-hard problems are really difficult to explain to people. VAS reachability is unimaginably harder than anything I've ever seen before, and yet can be explained to a fifth grader. That's why it's so cursed.
18
24
2
539
35,464
Why are so many people so quick to excuse LLMs generating poorly readable code? Is it not possible to recognize the usefulness and power of a technology while still being critical of deficiencies?
1
3
161
Yes of course we can work with the deficiency. Do you prefer to do that indefinitely or to have labs address it?
41