oh hello
Melbourne, Victoria
Joined September 2010
- Tweets11.4K
- Following738
- Followers416
- Likes21.5K
Josh retweeted
I recommend everyone upgrade Ruby immediately. This is a DoS and RCE vector if you accept user input into Regexp.new() ruby-lang.org/en/news/2026/1…
Josh retweeted
congrats, you vibe-coded your own software. you’re also the head of security now
yesterday, Railway was breached and thousands of vibe-coders (and companies) spent the day rotating every API key / password / token they had stored on there
then had to audit their entire tech stack and assess their exposure (i.e. maybe they weren’t exposed directly, but vendor A they use was)
same day, PHP (the language behind ~70% of websites) patched two bugs that could let a stranger take over your server without a password
if you are building your own software, all of that is on you:
> you need to stay up to date on the latest cybersecurity news
> you need to routinely audit and update your tech stack
> you need to spring into action to reduce exposure on days like yesterday
> you’re responsible and liable for your (and your customers’) data
or rather than vibe-coding your own storefront, you could pay shopify $39/mo and let 7,600 experts handle payments, fraud, PCI compliance, and all of the above for you
rather than vibe-coding your own newsletter platform, you could pay @beehiiv $49/mo and let 130 specialists handle deliverability, security, uptime, and everything else
AI made it easier to build your own software… but it also made it easier to break into it. there will be many more p0 exploits like yesterday moving forward
SaaS isn’t invulnerable, but at least those companies are staffed with senior engineering teams and security experts who handle issues like this for a living
Two critical vulnerabilities have been identified in PHP, a widely used programming language that powers websites and web applications. These vulnerabilities (CVE-2026-109618 and CVE-2026-109619) could allow malicious cyber actors to remotely execute code in PHP core, potentially compromising sensitive data and networks.
The FBI urges organizations to take the following steps now:
➡️ If an immediate update is not feasible it is recommended that organizations take the system offline until it can be fully mitigated.
➡️ Immediately update affected PHP installations following the guidance on php.net. Inventory where PHP runs in your environment, prioritizing internet-facing systems and vendor products that bundle PHP.
➡️ Monitor web servers for suspicious activity, including PHP processes unexpectedly launching command shells or downloading files. Ensure appropriate process-monitoring capabilities are in place to help identify potential exploitation.
➡️ If you find signs of compromise, isolate affected systems, preserve your web server and PHP logs, and report it to your local FBI field office or ic3.gov. Those reports help us identify the actors involved and warn other organizations before they become victims.
For the latest information, follow PHP’s announcement on php.net, CERT-EU Security Advisory 2026-018, and CISA’s Known Exploited Vulnerabilities catalog.
ALT Critical vulnerability alert for PHP Core with CVE-2026-109618 and CVE-2026-109619, CVSS score 9.2, urging immediate patch and threat hunting.
Josh retweeted
Some stats on #php2shell. In the 10 hours before it was publicly announced, across the @Cloudflare network we saw thousands of attempts to exploit it. That jumped to tens of thousands in the 45 minutes post announcement, hundreds of thousands within 2 hours, and millions now.
Josh retweeted
This is the most serious 48 hours in cybersecurity history.
"Black October" they're calling it
It seems every *hour* there's a new incident, that in a typical year, would be one of the biggest incidents of the *entire year*
SUMMARY OF THE LAST 48 HOURS:
YEAR-LEVEL INCIDENTS (In a typical year, these would be among the biggest incidents of the year):
1. PHP2Shell: a hole in the software behind ~70% of websites. Governments took sites offline.
2. eMudhra: the master keys that prove websites are real were cracked. One was the number 4.
3. California: data on every public school student since 2023 stolen, including disability status
4. Railway: hackers stole customers' secret keys from a hosting platform used by millions of developers
5. Zeabur: 321,810 accounts breached. The company is shutting down.
6. LiteLLM: an AI tool downloaded 3.4 million times a day was poisoned with a password stealer
7. GitLab: a 10-out-of-10 severity hole in every version since 2014
8. Cairn: criminals used an AI hacking tool to break into 100+ companies
9. Ahold Delhaize: hackers knocked the owner of Food Lion and Stop & Shop offline
10. Onamae: malware planted on up to 26,739 customer servers in Japan
MONTH-LEVEL INCIDENTS (In a typical month, these would be among the biggest incidents):
11. Signal: a bug could leave your mic and camera on after you hang up
12. NGINX: 3 critical holes in software running ~1/3 of websites
13. HTTP/3: a single connection can crash 7 of the 10 web server types tested
14. Plex: hackers breaking into home media servers
15. SafePay ransomware: FBI warns it has hit 600+ critical infrastructure organizations
16. Palo Alto: ransomware gangs breaking in through company firewalls
17. Bullhorn: platform used by 10,000+ companies knocked offline
18. Medi-Cal: 970,000 patients' data stolen
19. WordPress: hackers stole a plugin maker's keys and used them to plant backdoors on websites
20. AOMEI: backup software's update server handed out a backdoor
21. Meta Muse: Meta's AI agent exposed users' personal and business secrets
22. Tor: 333 relays hijacked to steal logins
23. TiDB Cloud: database company's cloud breached
Also serious:
24. State Bank of India: India's biggest bank confirms a data breach
25. OVHcloud: hackers slipped a password stealer into a cloud company's official tool
26. Siemens: a software hole being used to spread malware
27. North Korea: state hackers launch a new attack campaign on South Korea
Josh retweeted
the new jev image model can help you with one of the world's hardest questions, when is the toast done?
use a toast as each step of your Score.
Plan project with supergrok and Claude. Estimate 1/2 day of work. Tell Claude to go ahead. Rush to supermarket BC closes in 10 minutes. Come back home. Milestone 1 spike is completed
Lmao
Josh retweeted
I'd say ~75% of my issues with AI coding disappeared once I gave my agent the ability to verify its own work
set up super grok and now suddenly I have too many things I want to build because the cost of execution is close to the cost of new ideas but now planning is the bottleneck 😓
most of what we call review today will turn into risk scoring and continuous verification… the job now is to continuously make the system good enough that you don't need the diff to feel safe.
I've been thinking more about the this reply and I think I figured it out.
> how did you trust it before when humans wrote it?
you mostly didn't, not really. you trusted the tests, the fact that it was already in production somewhere similar, the author's track record, and the reality that most of the time the bug would show up in monitoring before it became a real problem. the diff was honestly a ritual more than a guarantee. we pretended like line-level review was a safety net because we didn't have better options and the volume was low enough that it felt manageable.
but agents just make the old ritual obvious. once you're landing thousands of changes a month the idea that someone is carefully reading every diff starts being theater.
trusting agent output is the same problem as trusting human output, just with higher throughput. I think the answer is still to stop treating the change as the thing you trust and start treating the environment as the thing you trust. if an agent can run the code, drive the ui, check the invariants, and get blocked by the same constraints a human would, the fact that it wrote the patch matters less.
every time you find yourself wanting to open the diff "just to make sure," that's signal that something in the system is still relying on human eyeballs instead of a hard constraint.
the code review as we know it is already on the way out. most of what we call review today will turn into risk scoring and continuous verification. low and medium risk work will land after agents and checks have already exercised it.
humans will still be involved, but not by reading patches. they'll just be deciding what outcomes matter, whats not allowed to break, and which constraints are worth making permanent.
the job now is to continuously make the system good enough that you don't need the diff to feel safe.
Josh retweeted
Dusting off this tweet from April because this gap in shared understanding of LLM capability is *widening*. It's now less "two groups of people speaking past each other" and more a sharp funnel.
- Napkin math somewhere around 6B people (~75% of the population) have barely come in contact with LLMs at all.
- Around 1-2B (~20%) are casual and infrequent users of free-tier ChatGPT-like products. This group sees derpy chatbots and treats them a bit like a better Google search, a writing aid, or etc. Maybe an agent tries to book you a flight. I have non-tech friends who (reasonably, imo) say they have not much use for it at all. Even many professionals outside of math&code are in this tier. For example, execs and many other functions spend a lot of their time talking to other people, so while they understand what is happening intellectually, it is still second-hand and a bit abstract.
- Now we get to professional use of frontier-grade LLMs in math&code. Somewhere around 20M people (0.2%) see first-hand that large, complex projects that used to take them weeks/months can now be completed by agents with a prompt. Building apps, copying apps, translating apps, decompiling apps from binaries... This has all happened very quickly and recently - less than 1 year ago, I was writing code manually by hand, typing memorized computer code commands into a code editor character by character, occasionally pressing Tab to autocomplete a little chunk of code.
- And finally we get to the ~5,000 people (~0.00006%) with access to frontier-grade systems internally. The external world has seen the preview. It looks like swarms of thousands of agents collaborating over weeks on software mega projects: minting zero days, running cyber attacks and defenses at machine speeds, discovering new science, advancing the frontier of mathematics. Things that would have taken top professionals in the industry years of work. Meanwhile, human review and comprehension are starting to fall behind. For example, people are still involved in the "archeology" of the OpenAI-HF incident from many months ago. Mathematicians may be poring over the 722 manuscripts on frontier mathematics for a while.
The funnel is driven by a combination of factors:
1. The impact scales with ambition, problem size, and horizon. A question with a paragraph answer barely stresses the system. You need a reservoir of big, difficult problems that you really care about. This aspect drives the consumer / professional dimension of the funnel.
2. The jaggedness of the system (which I have written about a lot separately). Capability peaks in domains that are digital, verifiable and economically valuable. This is because LLM capability emerges from reinforcement learning on verifiable rewards on a curated environment mixture driven by revenue potential. This aspect primarily drives the area (e.g. math&code) dimension of the funnel.
3. Access. Free-tier, paid-tier, internal.
So this is the weirdness of the moment. The general public has mostly not interacted with these systems. When they have, it looks like a derpy chatbot. The majority of professionals still see only a modest uplift. And a small sliver of professionals are experiencing the vertigo of the curve going vertical. And it is all happening at the same time.
Judging by my tl there is a growing gap in understanding of AI capability.
The first issue I think is around recency and tier of use. I think a lot of people tried the free tier of ChatGPT somewhere last year and allowed it to inform their views on AI a little too much. This is a group of reactions laughing at various quirks of the models, hallucinations, etc. Yes I also saw the viral videos of OpenAI's Advanced Voice mode fumbling simple queries like "should I drive or walk to the carwash". The thing is that these free and old/deprecated models don't reflect the capability in the latest round of state of the art agentic models of this year, especially OpenAI Codex and Claude Code.
But that brings me to the second issue. Even if people paid $200/month to use the state of the art models, a lot of the capabilities are relatively "peaky" in highly technical areas. Typical queries around search, writing, advice, etc. are *not* the domain that has made the most noticeable and dramatic strides in capability. Partly, this is due to the technical details of reinforcement learning and its use of verifiable rewards. But partly, it's also because these use cases are not sufficiently prioritized by the companies in their hillclimbing because they don't lead to as much $$$ value. The goldmines are elsewhere, and the focus comes along.
So that brings me to the second group of people, who *both* 1) pay for and use the state of the art frontier agentic models (OpenAI Codex / Claude Code) and 2) do so professionally in technical domains like programming, math and research. This group of people is subject to the highest amount of "AI Psychosis" because the recent improvements in these domains as of this year have been nothing short of staggering. When you hand a computer terminal to one of these models, you can now watch them melt programming problems that you'd normally expect to take days/weeks of work. It's this second group of people that assigns a much greater gravity to the capabilities, their slope, and various cyber-related repercussions.
TLDR the people in these two groups are speaking past each other. It really is simultaneously the case that OpenAI's free and I think slightly orphaned (?) "Advanced Voice Mode" will fumble the dumbest questions in your Instagram's reels and *at the same time*, OpenAI's highest-tier and paid Codex model will go off for 1 hour to coherently restructure an entire code base, or find and exploit vulnerabilities in computer systems. This part really works and has made dramatic strides because 2 properties: 1) these domains offer explicit reward functions that are verifiable meaning they are easily amenable to reinforcement learning training (e.g. unit tests passed yes or no, in contrast to writing, which is much harder to explicitly judge), but also 2) they are a lot more valuable in b2b settings, meaning that the biggest fraction of the team is focused on improving them. So here we are.
Josh retweeted
i love having a little family. i love cooking and the trampoline and reading harry potter. i love picking up lego. i love when my little toddler learns a new word or discovers kimchi is spicy. there is almost infinite reward in the banal & quotidian and its your job to get it out
Oh ffs
BREAKING:
5 hospitals have been closed for quarantines in Siberia due to a possible plague outbreak.
The 28-year-old lab technician Darya Shipilova, died yesterday after breaking a test tube with live Yersinia pestis the Irkutsk Anti-Plague Institute on September 25th while handling samples.
A mask mandate has now been introduced at the huge aluminum plant in Irkutsk.
Some U.S. politicians are now calling for commercial air travel to and from Russia to be stopped until the situation is clarified.
This first symptoms of pneumonic plague typically start about 4 to 7 days after exposure. 200 people are under observation.
Josh retweeted
jev class of models have no generative decoder loop, have bounded outputs, utilize parallel inference, & have extreme compression/efficiency incentives. this makes them perfect to run on device (this will happen soon).
at that point you can put judgment into basically every existing interaction like notifications, email, camera, keyboard, home screen, sensors, shortcuts, accessibility, etc. with zero marginal cost & it would be near instant (users would never feel the lag of an llm).
more specifically you can frame lots of problems that are potentially too expensive for an llm into decisions, stuff like:
should i surface this?
is this unusual?
does the user care?
which action is appropriate?
is this interruption worth it?
did their intent change?
should this ui adapt?
this basically adds a new computational programming primitive for ~free.
Josh retweeted
Here's a 45-second TL;DR on Jev.
I find the core idea beautifully simple, but the video made it really hard to understand.
Hope you find it helpful.
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
Josh retweeted
Liberals need to make the argument that immigration strengthens America CULTURALLY, not just economically.
And not because of "diversity", but because many immigrants are the type of people who built America and who keep it strong and dynamic.
Anti-immigration sentiment is sweeping the developed world.
Only liberal ideals can save us. Here's what that looks like:
noahpinion.blog/p/liberalism…
Josh retweeted
Hey folks,
I'm launching a website today: videogamescritic.com
It started as a little database I built for myself. I use Metacritic and OpenCritic a lot and they're clearly useful, but to me as a designer the user experience always felt a bit lacking.
Then the thing kept growing, and at some point I realized it had become genuinely useful for any gamer, not just me. So here it is, for everyone.
My main problem with the existing aggregators: Their scores tell you how the press felt about a game, not how players did.
That made sense 20 years ago. I cared a lot what GameSpot gave a game back then. But today we can actually measure how players felt, and to me that will always matter more than how the press did. On videogamescritic.com, players hold at least two-thirds of every score, by rule. Press still counts, it just can't outvote the people who bought the game.
The other big problem: Those sites rate a game once, at launch, and then never really touch it again.
But that's not how games work anymore. A game can launch rough and get patched shortly after, but on Metacritic and OpenCritic it carries that scarlet letter forever.
No Man's Sky is the obvious example: Famously rocky launch, and Hello Games has spent the years since relentlessly improving it. Its Metacritic score has never moved. On videogamescritic.com it launched at 50 and sits at 82 today, because scores here are LIVE: launch-era reviews fade out over time, and reviews written after a significant patch count the most. When player sentiment changes, the number changes. Simple.
And since press does still count, I ended up reviewing the reviewers too. Every outlet and creator gets a trust score based on one question: When they reviewed a game, did players end up agreeing? You can see who consistently calls games the way players see them, and who doesn't. Again, simple and fair.
And everything is transparent. Every game page shows exactly how its score was computed, and the full methodology is public. No editorial opinion, no magic number.
Anyway, I hope you get a kick out of it! Even just as a way to see the best games on each platform, find something you missed, or browse a site that isn't wallpapered in ads, I think it's already pretty useful.
Enjoy!👍❤️
