Tyler Willis (@tylerwillis) header
Tyler Willis (@tylerwillis) profile picture

technically, a member of staff || ceo of unsupervised

Seattle
Joined April 2007
Replying to @steipete
I think a lot of people don't get that this will become accessible as the cost of intelligence continues to fall. Only a few organizations can do this today, but the frontier will move and this style of automation will be commonly available to most in a few years.
21
10
2
397
96,292
Tyler Willis retweeted
I rebuilt the agent harness the way a browser runs web pages, and today I'm open-sourcing it! Affordance is a browser for models: you write an agent as an app, and it runs the loop and renders the exact tokens the model reads. (1/13)
3
3
1
22
908
There’s a good chance that swarms are one of the first products where multi-model harnesses have a structural advantage on cost and output quality. Labs can more easily reward coordination behavior in training (and that might be worth a lot), but the diverse opinions from different families might be worth more.
Swarms are quickly becoming one of the most important topics in Silicon Valley, sharing a few notes from recent conversations: - The naive view of swarms is that they’re a sum-of-parts method of parallelization / brute force, but we’re empirically seeing the “shape” of swarm intelligence be very different. - The way swarms achieve this is through a combination of parallelization, entropy, and specialization, leading to emergent solutions that aren’t covered by a small number of brilliant frontier models acting as central planners. - It’s also not obvious that the economics of swarms follow higher cost = better output. Swarms that mix and match frontier and non-frontier models with different reasoning levels may be more economical than a small number of frontier agents, and broadly, there should be a new Pareto curve here. - Swarms will be orders of magnitude more token-consumptive than agents, just as agents were compared to humans + models, further driving constraints in inference supply. I think it’s important to build intuition here so have been running some experiments that I’ll share soon. A great way to get started with swarms is the /swarms command in the new google anti-gravity. A few of the smartest thinkers on this topic are @alehandromz @jsnnsa @gokhan_egri @pirroh @amasad Love to hear other POVs on swarms if folks have them!!
1
2
221
Tyler Willis retweeted
Today I'm announcing Workers IO to solve the slowest part of engineering, verification! AI agents can now write years’ worth of code in days, but it still takes significant engineering effort to verify what’s being built. It’s impossible to predict all scenarios that could lead to a failure and that’s why most expensive failures are found in production. Workers IO runs millions of simulated production scenarios on your entire system, including dependencies. These simulations run inside a deterministic hypervisor which injects faults like network delays, disk errors, clock skew, and process failures to verify how your system handles these conditions. Deterministic execution allows us to systematically explore the state space of the system and perfectly reproduce failures. We are already helping customers do years worth of production hardening in days and build confidence in what they’re shipping. If you are working on systems that you can never afford to break, DM me and we can help you battle test them!
142
73
47
924
158,610
Tyler Willis retweeted
We're making wholesale storefronts free. Today, we launched Overjoy Storefront, a direct B2B ordering channel for consumer brands. Most brands currently pay for separate tools to manage their wholesale CRM, storefront, and orders. We believe all of this should live in one place and moving quickly toward owning the entire wholesale sales workflow.
Overjoy Storefront is now live. And it's free forever for brands signing up through November! Custom buyer pricing, net terms, rep accounts, and native integration with Overjoy CRM. Your entire wholesale business, in one place.
5
3
2
15
1,634
Time to bring back buzzfeed quizzes
i made a library of 34,220 persona neurons in qwen please help me find the weirdest ones: qwen-neurons-site.vercel.app…
1
7
595
RT @junhyeokkwak: Yesterday alone, we added $260K in net new ARR at Overjoy. We’re on track for another 27% MoM growth. Our demo volume h…
2
21
Whenever someone predicts a thing that will take a long time for ai to automate, there’s a good chance that’s actually the next thing ai will automate.
2
1
13
871
Reminds me of the time @theo gave design feedback to a model by simply saying “unacceptable.”
You’re all prompting wrong
1
7
396
We’re gonna end up with a lot less open web if we’re not careful.
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.
2
5
310
One cool thing about working at @every is I get paid to write about my pet Star Wars theories every.to/context-window/mini…
3
108
Seems like all the reactions to this are assuming either a completely-perfect simulated fruit fly matrix or that this is just a fruit-fly-suit that can be used to fly around in a game.
We placed the fruit fly brain inside Odyssey-3, a frontier world model. The lil’ guy is now flying around totally imagined, simulated worlds all by itself 🫶
3
415
I've been obsessed with the optchat architecture over the past week or two. This currently has under 100 stars -- I fully expect it will have thousands in November.
Source is out: github.com/VictorTaelin/Unii… I've been using Unii as my only agent for a month, and I couldn't be happier. I'm free from the torture of /compact, AGENTS.md, managing Claude sessions, and I can't overstate how empowering that is. I also keep receiving enthusiastic DMs of people who forked it to do their own thing, but I'm concerned that your AIs might get some tricky parts wrong, like caching and the exponentially decaying logs. So, please, ask it to study my codebase and fix anything that might be wrong in yours. I just want to make sure you get the right experience, because it is actually magical. People have ported it to OpenCode, Pi, and many other harnesses. If you did so, feel free to post it below. Also check UniiChat.com for my exact version!
1
1
282
A lot of what I've played with* here has been around using this in multiple different threads and having them "fuse" or share memories later. I haven't figured it out, and I'm not confident that it's possible to do it nicely, but I think it's a promising idea.
1
41
I’ve only been able to skim this so far, but it has now lived rent-free in my mind for hours.
3
10
794
Fears of vibe coding leading to a glut of sub-prime software seem wrong. The only real way to compete effectively now is to invest way more time and care into what you build. We're gonna get a lot more beautiful software and a lot more performant software.
9
1
12
1,034
Of course, there will be slop™, but when you get to pick what software to use, you'll get to pick wonderful options. And for the terrible stuff (and a lot of the good stuff), it's likely that it will mostly be an agent using it for you.
1
97
This is very exciting/encouraging
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works already-clean.goodfire.com/
3
7
299
Race the frontier.
2
4
319
Azlen has been on a tear with great explorations. Scroll back on @azlenelza to see what I mean.
Breaking apart a book
1
5
561
Just brought my 11in MacBook Air back to life with Omarchy. From boot to configured in less than 10 minutes.
1
3
336