#GAAD (Global Accessibility Awareness Day) with a 200M+ social media reach AIMAC, an AI benchmark for A11y @A11yGenAI my podcast @a11yaudits
Henderson, NV
Joined February 2009
- Tweets43.2K
- Following6.1K
- Followers6.5K
- Likes65K
Pinned Tweet
I'm excited to announce that The White House will be doing a Global Accessibility Awareness Day (#GAAD) Event and Town Hall tomorrow at 10AM EST.
Links to Livestream and Access Streamtext in the comments.
Thanks @WhiteHouseOPE46!
From blog post to The White House. 🤯
Joe Devon retweeted
The Ocean Cleanup is now $40M/year short of what’s needed to stop up to 1/3rd of all plastic flowing into the oceans by 2030.
That’s less than 1% of what America spends on Halloween decorations each year.
This seems legit and is incredible. @skdh RTed it, which is more than enough for me.
A company in India will use AI to custom design a personal vaccine for cancer patients for $300k. Costs will come down with volume.
Let’s root for them to succeed in this endeavor. Especially poignant for families, including mine, who have dealt with a family member battling or dying from this horrible disease.
God bless you and GO INDIA! 🇮🇳 Impressive development.
I don't think people realise how big a deal this is.
If you have cancer, or somebody you know has cancer, there is a company in Hyderabad that wants to build a personalised cancer vaccine for you.
Try approaching any big pharma company and asking for a personal vaccine. If they even reply to your email, this is the response you'll get: we don't do that here.
PopVax, on the other hand, takes your tumour sequencing data (or a sequence you've designed), designs candidate neoantigen vaccine sequences with their AI pipeline, produces the mRNA, encapsulates it in their LNP formulation, fills vials, and runs QC, all in 8 to 10 weeks. They’re publicly committing to capacity for 100+ of these over the next 12 months and asking people to email them their cancer type/stage and sequencing details.
What a time to be alive.
Joe Devon retweeted
This is extremely important for all to understand!
I've been saying the same and walking the walk: we replaced 2000+ third party (incl. transitive) supply chain dependencies, and now we have basically zero.
A software company should control its software.
Fork your dependencies, trim them to only your use case, never update unless it breaks for your users. I’ve been vocal about this for 10+ years. I’ve always said that updating is way riskier than latent bugs (which can be tracked and CVEs monitored).
If you are updating a dependency, it’s on you to analyze every single commit in the full transitive set of dependencies. If you dont see anything compelling, dont update!
I remember at HashiCorp once in awhile an engineer would try to update a dep or replace a DIY lib with an external one and id always ask “show me the commit we need.” Dont update for the sake of it.
Feeling pretty swell about this mentality with all the supply chain attacks happening.
Wow this is so cool.
Joe Devon retweeted
"I don't think minds are made—either biologically or by AI—I think what you make are interfaces."
It was a delight chatting with @drmichaellevin for my new podcast, Human Potential Lab. We nerded out about where genius comes from. Watch the full episode: youtube.com/watch?v=XTL9PxDm…
I see everything.
Everything.
What. A. Masterpiece.
Well done @nathanfielder. 👏
If you’ve ever dealt with S tier gaslighters in your life, you will meet several in this movie.
Reddit is ending anonymous login.
So if you want to read the "NSFW: Furry back scratching with Coca Cola bottle caps" subreddit or similar, you will need to at least have an account.
nitter.cf/i/trending/21159167553…
Joe Devon retweeted
Meet NVIDIA Nemotron 3.5 Transcribe 🎙️
An open-weight, promptable 30B MoE model with ~4B active parameters that brings transcription in 40 languages, speaker attribution, translation, contextual prompting and audio understanding into one checkpoint.
Watch it keep track of who said what when everyone’s talking at once.
1/ 🧵
Joe Devon retweeted
An insane insane insane amount of effort went into testing this new algorithm over the last few months and we're incredibly proud of where it landed. Check out the blog for more information!
We just shipped our biggest improvement to our provider router since the founding of OpenRouter.
Before, we prioritized mostly based on price. The new router now rewards the best quality inference for the actual cost of a request, so requests stay cheap but are also fast and reliable.
Providers that deliver more value automatically win more traffic, so every provider has a reason to keep improving. And every time a provider gets better, so does your inference on OpenRouter.
When EF Hutton, or Shreya Shankar talks, everyone listens.
Finally something unique about fixing AI Slop, by the Queen of Evals.
That means this list has probably been eval’d and works.
First thing I will be doing when back at laptop is improve my unslop skill with this checklist.
Thank you Shreya!
This.
Deep work.
Built a *nix kernel for research.
I love this effort so much.
The world has changed fundamentally in the last couple of years, but all we see are glimpses of what's possible.
It will take years for the changes to trickle out.
Work like this is the glimpse.
People are missing the point of Ginux.
I didn’t create another kernel to compete with Linux. That is idiotic!
In fact I consider the code disposable. Think about that…
Ginux wasn’t the product. The development process was the experiment. The kernel was the evidence.
Could AI, directed by an experienced engineer, could build and validate an extraordinarily complex software system.
In just 11 days, I produced an independent kernel capable of booting Rocky Linux 10 with systemd.
That was the experiment. And while Ginux isn’t a complete Linux aligned kernel, the experiment is complete.
The research paper documents the methodology, results, and limitations.
Since publishing, people have started finding bugs and security issues. That’s fantastic! Independent review is exactly how complex software gets better.
There has already been an initial security review, and there’s a remediation plan sitting in context that I’m considering implementing.
So the question is, since the experiment is complete, do I even do anything further here?
Maybe there is an interesting experiment in measuring how efficiently human-directed AI can fix the problems external reviewers discover?
How much time does it take to reproduce an issue, identify the root cause, implement a fix, and validate the result? How much human intervention is actually required?
That’s potentially an even more interesting situation than the original 11-day experiment.
Because the real question was never whether AI could replace Linux. Even if it can replace some aspects of its functionality, it can’t replicate the community, standards, collaboration, alignment, or stability.
If you still believe that Ginux was designed to be a Linux replacement, you should read the actual paper.
zenodo.org/records/23682548
I love the ethos of a ban for using ai for replies a single time, but sometimes people will sound like AI as a joke. Sometimes you have false positives.
But the people who cause the worst problems are those that use ai replies at scale and that’s easy to detect manually so it’s probably even easier to detect with AI.
X should have a zero tolerance policy for AI replies
Like you do one and your account is suspended for life
Replies are a nightmare now and it's never ever been this bad
AI replies are an an existential threat to the future of this platform, because people will simply stop using it if the vast majority of replies are AI bots
Joe Devon retweeted
Fascinating case study of the sharpest edge of what you can do with agentic engineering in the Fable era - my eyebrows raised at least a dozen times while reading this
Wow what a crazy story. LLMs + robot + mouse + brain surgery. Gabriel is calling this procedure a lab in the loop.
Account Followed.
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...]
Joe Devon retweeted
1/ Excited to announce our new company Particle
This is our first step in a series of models focused on developers.
With particle-1, we focused on token efficiency and managed to keep parity in performance while using ~45% fewer tokens.
Live now: particle.ai
Joe Devon retweeted
80% of Anthropic's revenue comes from API consumption pricing
This is their soft spot. DeepSeek costs 20x less than Opus and works for 70% of the tasks already
Companies are massively overspending if they use expensive tokens for the majority of simple tasks
Joe Devon retweeted
I did say we'd be releasing our models periodically...introducing Gemini 4 Carbon, our next frontier model.
Following Argon's release just two weeks ago, Carbon shows the pace at which we’re pushing the frontier, it ranks #1 on the recently updated v5.0 of @ArtificialAnlys Intelligence Index at 59. It also leads across key cybersecurity evaluations scoring 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching.
We’re seeing significant improvements in long-horizon tasks even compared to Argon, and we’ve strengthened frontier safeguards.
As we did with Argon, we’re going to roll it out gradually + responsibly. Some trusted cyber defenders will get access starting this week. We'll be expanding it to more partners through our Fairwind Program, ahead of wider availability soon, so hang tight!
