#GAAD (Global Accessibility Awareness Day) 200M+ social media reach AIMAC, an AI benchmark for A11y @A11yGenAI my podcast @a11yaudits
Henderson, NV
Joined February 2009
- Tweets42.8K
- Following6.1K
- Followers6.5K
- Likes64.3K
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. 🤯
Sad to see that one of my favorite yet underused features by @OpenAI is silently going away.
Custom GPTs will be sunset December 11 apparently. People are being asked to convert them to plugins.
No thanks. I used this everytime I post an image. I made it myself. I hate plugins.
Thankfully @huggingface released their own open source Custom GPTs. Alas, I can't find the page @ClementDelangue. Was it taken down?
ALT VisionScribe’s ChatGPT page shows a prompt to switch to GPT-5.5 Thinking, example alt-text requests, and a notice to migrate GPTs to plugins by December 11.
Claude tip:
Now that Claude Code has finally implemented AGENTS.md support, they've really cooked in their implementation.
I recommend setting it to claude-md-and-agents.md. This way if you have a Claude-specific instruction, put it in CLAUDE.md. Otherwise it goes into AGENTS.md.
While they could improve the look and feel of that configuration, I don't care. I just deleted a bunch of symlinks. So much easier now.
Thanks to @bcherny, @_catwu and the team. You heard our feedback. Well done.
ALT A terminal-style dialog titled “Project instructions” lists how CLAUDE.md and AGENTS.md files are loaded. Visible text: "claude-md": CLAUDE.md only, loaded by the engine as today. "claude-md-or-agents-md" (default): a project with no CLAUDE.md of its own gets its AGENTS.md files instead, loaded exactly where and how CLAUDE.md would be. "claude-md-and-agents-md": AGENTS.md files are loaded beside CLAUDE.md (a file CLAUDE.md already imports or links to is not loaded twice). "managed-only": the project's and your own instruction files are dropped; the organization's managed CLAUDE.md and memory stay. Options shown: 1. claude-md 2. claude-md-or-agents-md 3. claude-md-and-agents-md, selected 4. managed-only Footer text: Enter to confirm · Esc to cancel.
Correct
In the obsession with AI, too many founders, leaders, managers seem to have forgotten that an engineer with motivation + drive + focus + AI >> hundreds of AI agents without any of those
The best products and companies are built by people and teams like that, and will continue to be so
Ask yourself, again, why the AI labs hire the best of the best (some of the most driven people in the industry; founders, CTOs etc) if people aren't making the differrence when everyone has access to similar tools already
We are on the bleeding edge and it's pricey. Subagents and loops and goals are great experiments. But they are not robust or reliable. One day you build a massive application overnight and your usage doesn't get used up.
The next day one agent launches 15 subagents who keep adding processes on top of processes and you have to throw out that day's work or undo it. (My day today)
And worst of all, you'll catch some of them in a loop literally doing nothing.
So now it's hard to trust subagents. Hard to trust loops.
We know they'll figure it out one day. But that day is NOT here yet.
LOL @MartinDevon tried to /bro me (unslop on grok. I think?)
Me: That won't work because I'm literally your bro so everything I say is bro talk.
I made an enormous mistake...
Remember when I told you I spent ~$4 bucks for 300M tokens on @deepseek_ai / @pidotdev for the last week?
Mea Culpa. I miscalculated the numbers. I didn't realize that my @OpenRouter token version of DeepSeek was in the rotation too. It completely changes the equation.
I can't find the number of tokens, but I also spent $0.00253 for Deep over OpenRouter.
I hope you accept this apology for misleading you.
Joe Devon retweeted
JEV actually doesn't even run deterministically
Exact same prompts give you different probabilities when you run it multiple times
The ORDER of the choices DRASTICALLY changes the output probs
I am more and more confused by the their "no-hallucination" claim
Guys it’s a misconception that Jev can’t hallucinate
It’s a non-generative LM ie it models probabilities of things given language as context
The guarantee is the type-safety, and that answers are deterministic - ie no stochastic sampling. The deterministic answers can still be confidently wrong!
It can still hallucinate if the training data on the topic was under cooked
I was wondering why Claude Code and Codex seemed so slow as I started using codex again after it finally reset and claude because it resets at 9pm and I want to use very drop of the plan.
Then it hit me! Because @pidotdev and @deepseek_ai are so fast and the traces scroll by you at light speed!
So much of our experience is related to our harnesses and workflows and we don't even realize how it affects us.
I support this so much. But I’ll be honest, I don’t know where to begin with 3M models to choose from.
I struggle to navigate Hugging Face. When I ask the clankers they seem to struggle as well. They only know about the very top models and don’t seem to do HF research well.
I wonder if any HF users have tips for me?
Please tag anyone who tends to be responsive that you think is a HF user.
@daniel_mac8 @NoemiTitarenco are you HF users?
What I like about Jev: for years, large generalist models sucked almost all the oxygen out of AI.
But there are huge opportunities in much more specialized and customized models, built for specific tasks and languages and as a result, orders of magnitude cheaper, faster and more optimized. There are 3 million of them publicly available on @huggingface.
Let's build a much more diverse AI ecosystem!
Joe Devon retweeted
What I like about Jev: for years, large generalist models sucked almost all the oxygen out of AI.
But there are huge opportunities in much more specialized and customized models, built for specific tasks and languages and as a result, orders of magnitude cheaper, faster and more optimized. There are 3 million of them publicly available on @huggingface.
Let's build a much more diverse AI ecosystem!
Joe Devon retweeted
A personal update! Last Friday I wrapped up my time at OpenAI, after an incredible and unforgettable ride. I've spent the last couple of months helping to transition the design team and I know they are in great hands. I'm leaving feeling more bullish on the future of the company, its leadership and the team than ever and extremely thankful for my time there.
Getting to lead the design of ChatGPT, Codex and all of our other products and experiments along the way has been a true honor. Even more rewarding has been the chance to build a team of the best and kindest bunch of designers I've ever worked with. I will deeply miss being in the trenches with you all and look forward to seeing you continue to sweat the details and to bring agi to the world. There's so much good stuff in the hopper, I can't wait to see it come to life.
OpenAI truly is nothing without its people, and its people rock. Thank you all for making it such a special place.
As for what's next, I'm excited to link up with @wjosephflynn and @turtlesoupy to try something new. Hopefully we can share more soon!
Today will be a twist.
Been using only Claude and DeepSeek for a week w/ Codex at 0%.
Codex is now at 100%. Fable is at 0% and only 7% left for Claude.
But at 9PM Claude resets and will be playing with the 3 models for a whole week.
Productivity should go through the roof
Joe Devon retweeted
Just shipped Jev-powered search in pi-mcp-adapter. Describe the job (not the tool name) and it can instantly find the right MCP tool. When I tested across 95 tools, Jev found the expected result on first attempt 91% of the time (and much faster) vs just 45% for regular search.
github.com/nicobailon/pi-mcp…
Claude:
"The 490 furniture rows from the 01:32 poll:
2D Delete the rows whose links sit in furniture on their page today...
2L Leave them for the candidate filter to judge once"
Me: "the furniture rule?"
Claude: "You've reached your Fable limit. Run /usage-credits to continue or switch models with /model."
Me: "/model Opus 5 (1M)
Claude: ""The furniture rule" is a label I made up, and I shouldn't have used it without defining it. It means your 1B: the page reader skips links inside navigation, header, footer and sidebars unless an article wraps them. I'll call it that from here on."