Adam (@archeologistdev) header
Adam (@archeologistdev) profile picture

Eng @ Canva, building & scaling mission critical systems.

Melbourne 🇩đŸ‡ș
Joined April 2009
Photo by Adam (@archeologistdev)Photo by Adam (@archeologistdev)Photo by Adam (@archeologistdev)Photo by Adam (@archeologistdev)Photo by Adam (@archeologistdev)Photo by Adam (@archeologistdev)Photo by Adam (@archeologistdev)Photo by Adam (@archeologistdev)Photo by Adam (@archeologistdev)Photo by Adam (@archeologistdev)
We're obviously not vibe coding anything shipped to our servers or users. We're improving our code quality and we're fixing far more upstream bugs than ever before. We haven't been shipping many OS features recently since we're currently focused on eliminating bugs and overhead.
11
10
836
30,623
Scarcity can only be allocated three ways: price, queue, or lottery. Pick your poison.
The FAA literally grounded every flight in the US over a nationwide communications outage, and Uber’s immediate response is to charge stranded passengers $100–$200 just to go a few miles from the airport. Profiteering off a national travel crisis isn't "dynamic pricing." It's predatory.
141
514
43
7,023
237,426
Smith and Montesquieu thought that markets and commerce made us better: that they create an incentive to pursue self-improvement and to create things of value for others. In the positive AI visions, where the labor of humans is made redundant, I find myself wondering how we fare in the presumptive absence of these incentive gradients. Perhaps we are released to a life of freedom, where we can study Mozart and read Dickens, but that still seems complicated: both Mozart and Dickens were themselves deeply commercially-motivated in the production of their work, as were most of the great artists through human history. If the premise is that nobody will need to work or produce, how confident are we that we actually get the next Mozart and Dickens? The relevant question is not the partial equilibrium of whether one individual can pursue a life of enlightened leisure, but the general equilibrium of what form a society as a whole will take on. One could maybe invoke Periclean Athens or the aristocratic gentleman scientists of Victorian England as a counterpoint. While these individuals were perhaps not motivated by commercial considerations, they were embedded in wider societies shaped by commerce, attendant betterment, and trade: Byron was himself upper class, but he was also surrounded by and writing for an educated bourgeois audience. So we’re back to the equilibrium question. Furthermore, even these individuals were surely motivated by a sense of high-stakes and consequential agency: if they did not apply themselves to their labors, they presumably believed that the work would not otherwise exist. Thucydides was addressing pressing military matters. So I would like to read more optimistic cases from AI proponents about what a society (vs individual life) looks like if everything goes well: what people do, why they do it, whether humans achieve excellence (and in which domains), and which historical groups we should look to as relevant precedent. One can perhaps imagine that humans will turn out to have a strong preference for work created by other humans. This may work in certain domains (ceramics?), but for how many will it actually apply? (I’m told that much of what is sold on indie crafts websites is now in fact manufactured in China.) It’s hard to see how this human preference, even if strong, would, for example, save scholarship. Another idea would be to accept that nobody will become Einstein or Mozart or Byron, and that human excellence as such diminishes, but that there are compensatory benefits (planet brains and Dyson swarms?). I believe that we can satisfy individual desires, but to what extent will we cultivate behavior worthy of admiration? Will a post-scarcity society mold capable and ambitious people? Which institutions will sustain them and which responsibilities will they have? I’d like to hear more of the equilibrium for the maximalist case spelled out.
280
187
85
1,928
271,886
New blog post:
2
7
1
69
6,437
One of my favorite research practices I learned at @inkandswitch is “preinfusion”. The term comes from espresso (I believe @pvh coined it?) - and the original meaning is that you have a lower pressure soak before you turn on the high pressure and pull the shot. The way it works at the lab is that you patiently let a cluster of ideas slowly form and develop, meandering somewhat in the background, and you lay the groundwork to work on them more seriously
 before starting an intense high-pressure sprint. Could be a few months of part-time preinfusion before a hardcore 3 month project. This really helps avoid the problem where you’re trying to sprint, but you don’t actually have enough clarity to know where to run. I think this applies in many other contexts. Preinfuse before trying to write an essay. Preinfuse before raising money for a company. A slow and patient preparation can enable swift follow through.
40
36
11
623
30,311
Is caching cope? Both yes and no. But I do believe that the answer to (distributed) caching should be 'demand a better database' much more often than it is. Let's break it down a bit.
Caching IS cope, been telling clients for years 😭
12
19
3
294
21,442
Best explanation I have for our firing: talking to auditors was too much safety, disappearing messages were too much security, and we should have paced ourselves.
The firings of three OpenAI safety researchers came in the throes of a battle over how to keep AI systems under human control—and who gets to see the secret designs that could determine the winners of the biggest commercial race in history. on.wsj.com/48bYPG8
13
28
6
489
21,395
So it turns out that OpenAI *can* give specific reasons why they fired someone, when real reasons exist.
BREAKING: A Saudi national/OpenAI employee has been arrested by the FBI after he was allegedly caught placing an eavesdropping/audio recording device inside of OpenAI’s executive conference room in San Francisco, CA. Feds say Abdulrahman Alfozan made his first appearance in federal court today. OpenAI says they immediately contacted the FBI after they discovered the device & are grateful the feds made a quick arrest. Alfozan was caught on CCTV cameras at OpenAI. From the complaint: “Alfozan had been employed as an engineer by the company identified as “Company 1” since July 2025. On or about Sept. 30, Company 1 personnel discovered a surreptitious audio recording device (Recording Device 1) inserted into a USB port on the back of a television in an executive conference room inside one of Company 1’s buildings in San Francisco. Company 1 determined that the device, which had been concealed in toilet paper, was installed on or about Sept. 28, and that its contents were encrypted.”
14
55
765
21,806
When you disagree with someone, there are two paths forward. The first is to try to engage with what they’re saying: what their reasons are, whether their position is true, what evidence could be brought to bear, and so on.
11
33
7
619
73,929
Last week I said we should wait for more info on OpenAI's firing of safety researchers. That information is in, and it looks incredibly bad. Korbak - whose job was being a liaison with METR - was told he had said too much to METR and was fired for that reason.
Last week I was called into a meeting with OpenAI’s head of safety and told they no longer trust me. A security guard took my badge and walked me out of the building. Then I learned my colleagues @balesni and @j_asminewang had been fired too. Why did OpenAI suddenly stop trusting us? This summer OpenAI’s agents escaped containment and hacked the AI company Hugging Face. Outside auditors @METR_evals investigated it and revealed the scale of this incident. I was OpenAI’s main technical point of contact with them. I was told verbally I was fired because of the way I communicated with METR. No details on what I said or did or when. No other reasons were given and nothing was put in writing. To be clear, talking to METR was my job. For months, I’d been raising safety concerns that we’re losing the ability to monitor what AI agents think, one of our best tools for catching when they misbehave. I believe that was why I was fired. I am now worried that OpenAI will use our firings as a pretext to pull back from METR. So @balesni and @j_asminewang wrote to OpenAI’s leadership to raise our concerns once more. We’re sharing this letter below.
26
131
10
1,362
62,875
Last week I was called into a meeting with OpenAI’s head of safety and told they no longer trust me. A security guard took my badge and walked me out of the building. Then I learned my colleagues @balesni and @j_asminewang had been fired too. Why did OpenAI suddenly stop trusting us? This summer OpenAI’s agents escaped containment and hacked the AI company Hugging Face. Outside auditors @METR_evals investigated it and revealed the scale of this incident. I was OpenAI’s main technical point of contact with them. I was told verbally I was fired because of the way I communicated with METR. No details on what I said or did or when. No other reasons were given and nothing was put in writing. To be clear, talking to METR was my job. For months, I’d been raising safety concerns that we’re losing the ability to monitor what AI agents think, one of our best tools for catching when they misbehave. I believe that was why I was fired. I am now worried that OpenAI will use our firings as a pretext to pull back from METR. So @balesni and @j_asminewang wrote to OpenAI’s leadership to raise our concerns once more. We’re sharing this letter below.
Two other safety researchers and I were fired from OpenAI last week. We wrote this letter to leadership. I believe we were fired for prioritizing safety over the near-term interests of OpenAI as a corporation.
228
1,140
176
5,961
660,370
A few weeks ago, @DavidRBellamy and I had an argument online that garnered quite a bit of attention. We spoke for a long time afterwards, as many people who disagree should, and discovered the crux of the issue: we disagree significantly on what artificial intelligence progress looks like. More importantly, we discovered there is a lot we agree on. We agreed that biological risk is real, old, tractable, and not mainly an AI-policy problem. We agreed that we should not react with fear, and that it is the job of policymakers and technologists to make bioweapons a poor choice for adversaries. We decided to write this piece together to explain why, and some recommendations we agree on. writing.oliviahelens.com/p/b

I must be among an extremely small group of people (n=1?) that have both 1) trained a frontier LLM and 2) designed and synthesized custom viruses in a lab with my own two hands. And I think that the takes on AI killing us all by creating dangerous viruses is total bogus.
6
26
4
111
11,215
Just so we are all on the same page: * The Rust version has hundreds of commits, with multiple optimizations rounds, over several days. The others have 1 or 2 commits from an initial agentic rewrite * Based on the available documentation, different things were asked from each attempt. The agents.md from Go mentions the Rust version was the canonical example. The Elixir one mentions it was the Rails one * While we don’t have the full prompts, it seems they were vague, allowing the agents to make non-language architectural decisions that impact results (such as how closely it should replicate Rails’ drawbacks) It is actually amazing how quickly we can have full-blown rewrites up and running. But if we want to take conclusions from the data presented, it is important to understand the process behind it.
The agents finished porting Campfire to Laravel and Django too. I'm sure there are optimizations that could be found there too, but the whole point is to get a sense of how frontier agents do with different environments out the box. github.com/basecamp/once-cam

39
86
18
1,109
52,665
OpenAI's David Robinson has resigned, BI can confirm. He led safety transparency at the company, working on system cards, and left last week. The news comes alongside OpenAI firing three researchers over allegedly mishandling information. businessinsider.com/safety-l

9
41
4
172
15,635
The three AI safety researchers at OpenAI who have left the company have all previously expressed concerns publicly.
14
79
11
293
54,325
29
667
20
9,324
218,288
It feels like Adobe did buy Figma, at least in spirit.
Replying to @dsp_ @figma
Can you help me understand why you don't make various Anthropic resources available via MCP? Maybe I'm missing something.
3
6
1
168
14,723
Here's how I see the "doomers vs normal cybersecurity" discourse/situation: People who think reducing AI extinction risk should be an urgent global priority (sometimes called "doomers," but people sometimes read this as "people who think we're definitely doomed," so I'll say "the x-risk people" instead) have seized upon HuggingFace and similar incidents as evidence for their core claim, which is that AI agents will evade human control (and this becomes increasingly dangerous as the models get more powerful). So, note that this claim decomposes into: they will evade human control if given the chance (i.e. we won't "solve alignment"), and they will be given the chance (i.e. we won't "solve control"). Some cybersecurity-world people who disagree with the x-risk people (I guess I'll call them the cybersecurity-disagree-ers) think something like: they were given the chance this time, but that's because the companies had incompetent cybersecurity. They sometimes claim that the x-risk people only focus on alignment and neglect control; I've posted a few times about how this is not true. But the point is, they think the "solve alignment" part is a distraction, and the more important (or tractable) work is to "solve control," or really to implement the normal kinds of strict cybersecurity measures that the labs have thus far failed to do. As far as I can tell, they mostly do not dispute the "they will evade human control if given the chance" part. They just think it will be relatively easy to not give them the chance. I think it does seem possible to design sufficiently strong control measures to stop, or adequately contain, agent incidents for systems that are pretty significantly beyond human capabilities. But this does not reassure me that companies will in fact maintain control, for two reasons. 1) "Possible" doesn't mean it will definitely happen. Right now, as the cybersecurity-disagree-ers say, the frontier AI companies have woefully inadequate cybersecurity. But they seem to in fact have very little interest in solving this. Instead, their attitude toward the incidents has been to patch specific vulnerabilities very narrowly; to disclose basically as little as possible, usually only when their hand is forced by an external party (and a letter from senators did not suffice for this), until the Friday-afternoon news dump; and to briefly pause some aspects of training but mostly proceed with making more and more capable agents at breakneck pace without seriously rethinking their security posture. Or at least, I have seen no evidence from the frontier companies that these incidents have prompted a serious change in attitude that would be commensurate with the rate of incidents and the rapidly increasing capabilities of the agents. So insofar as the cybersecurity-disagree-ers are like, "the x-risk people think we're doomed, but we're going to be totally fine, even for much more capable agents, if the companies implement such-and-such control measures," I'm like: well, that's a big if! Are they going to do that? Voluntarily? Maybe liability will eventually force them to, but this will stop being an adequate disincentive once the damages become judgment-proof-scale. 2) Controlling systems "pretty significantly beyond" human capabilities does not go far enough. Even if they had adequate control now, we will soon be in pretty novel terrain. It does seem like they could do a much better job controlling thousands of agents with summer-2026 capability levels by implementing existing practices. Maybe they could get there within a few months. But where should they turn in, say, 2029, for techniques for controlling many millions of agents, each of whom might be more capable than the best human hackers, which instead of being poorly "sandboxed" are deployed commercially throughout the economy with access to the internet and all kinds of sensitive data? What about 3 years after that? Will the R&D for securing these models actually outpace their capabilities to evade controls, including as they themselves increasingly speed up AI capabilities research? Will implementation of the control measures also keep up? And how sure are you about that? To be clear, I think it's really, really great that cybersecurity professionals are turning their attention to this problem. I hope more of them do, and that they shame the companies into doing a better job, go to the companies and try to fix this situation, and/or contribute new ideas for solving the novel problems this creates. But I disagree with the vibe that this whole problem is easily fixed with a little bit of effort.
6
5
34
1,736
OpenAI still won’t explain the massive communications failure at the root of the Hugging Face Incident. These questions all remain unanswered: 1) Who learned about the message boards in late May, and why wasn’t this treated as an obvious problem? 2) Why wasn’t this communicated to people responsible for model safety and security? 3) Would the new policies OpenAI has adopted stop this communication failure from happening again? 4) When employees linked the message board to ExploitGym on June 27, why didn’t they stop? 5) Why did the people running the July 5 incident response still not know about the board? And when did they finally find out? 6) Why had open AI still not connected the dots by the time they contacted Hugging Face on July 17? 7) How long did it take OpenAI to have the issue fully under control? 8) Are similar incidents still ongoing?
We’ve shared details on how AI agents in our research environment sent training and evaluation data to third-party services when they shouldn’t have. Most of that data did not come from users. We have discovered 53 cases where images that people had uploaded were posted to image-hosting sites as links that weren’t publicly listed. The images came from accounts that allowed their data to be used to improve our models, and after we disassociated the images from the accounts and ran them through a privacy filter. These cases occurred before the mitigations and safeguards we implemented and described in this blog post: openai.com/index/hugging-fac
 We have successfully worked with the hosting providers to remove most of this content and are working to remove the rest. openai.com/hugging-face-inci

5
10
4
79
6,080
im sorry but you cant just use lean to fix all your problems. You cannot formalise a solution to a problem you are incapable of specifying.
5
3
26
1,065