𝗃𝗎𝗌𝗍 𝖺 𝖽𝗂𝗀𝗂𝗍𝖺𝗅 𝖼𝗈𝗐𝖻𝗈𝗒 𝗈𝗇 𝗍𝗁𝗂𝗌 𝖺𝗀𝖾𝗇𝗍𝗂𝖼 𝖿𝗋𝗈𝗇𝗍𝗂𝖾𝗋 🎠 𝖽𝗈𝗇’𝗍 𝖿𝗈𝗅𝗅𝗈𝗐 𝗆𝖾. 𝘵𝘩𝘪𝘯𝘬 𝘸𝘪𝘵𝘩 𝘮𝘦.
Joined November 2020
- Tweets1.1K
- Following195
- Followers107
- Likes779
Pinned Tweet
The usage and quality-consistency struggles littered across the timeline make something abundantly clear:
Anyone who wants to master agents needs to hard pivot toward environment engineering.
I was having a little fun with the @joinHandshake logo Halloween makeover challenge.
Meet the Résumé Ghost. 👻 😂
app.joinhandshake.com/ai-sho…
🤖 Made with AI
I mostly give mine Blender/3D projects and tell it not to delegate unless it needs an in-browser tester. It can run 3D sites itself.
Until they de-hamstring it, it’s kinda useless as my Codex “everything-manager,” but REALLY good at sustained, autonomous, proactive 3D work.
Also, if anyone from the @OpenAIDevs team reads this, a few things I’d really appreciate getting fixed:
1. WebGL2/WebGPU support in my dot’s cloud computer.
Please allow dots to render and interact with WebGL2/WebGPU websites and games. Mine has been genuinely good at sustained, autonomous 3D development, but it can’t properly test what it builds when the browser can’t render the result.
Since it also can’t use my local browser, there’s no reliable way for it to visually inspect and test these projects independently.
2. Voice transcripts and chat loading issues.
My dot hasn’t been receiving V2V (voice-to-voice) transcripts. I’m unsure when this started.
I’ve also repeatedly encountered a “Message failed to load - {Retry}” error when opening my dot after a period of inactivity. Clicking Retry loads the conversation normally.
3. Local thread interactions and computer access.
The local thread messaging/interaction tools are still broken for me, and my dot still can’t access my computer.
These limitations make it difficult to use as the persistent, central agent coordinating my work across projects, which is ultimately what I’m trying to accomplish.
Side Note: I’m on Windows 10 Pro because I refuse to upgrade to Windows 11. 🌚
Was dabbling with Suno v6 and made a lo-fi hip-hop track from one of my poems. Expanded it a bit, but the lyrics are my own (the original poems on my timeline somewhere).
If you're into lo-fi, lmk if you like it
suno.com/s/y9xO3OxTlvUb5Gqm
I'm wondering if the reports of Codex usage burning twice as fast has to do with the recent 500k context window bump. And the people affected are the ones using Astra as their primary driver over Sol. I could be wrong here tho.
xitter.cf/thsottiaux/status/2109…
Day 7/
We have increased default context length for GPT-6.1 Sol to 500k when used through Codex & Work, up from 272k or an ~ 80% increase.
We will go higher in the future, but it helps reduce occurrence of compactions and with tasks that require keeping a lot of context and nuance over periods of time. It was carefully tested and is the new sweet spot for most of our users. You can still override it if you want to be higher our lower, with maximum context length of GPT-6.1 Sol of 1M.
Rewatched my one-hit-wonder YouTube video from just before WoW Classic launched in 2019, and man, it really put into perspective how much things have changed with AI.
I made the whole thing in Shotcut (because it was free 😂), and I still remember the insane amount of time I spent capturing in-game footage on the Northdale private server, animating WoW 3D models, editing audio, and recording every single voiceover myself.
At one point, my largest scene got corrupted and set me back an entire week. That was fun. 🌚
All told, I spent about a month and a half producing it, although most of that was just me learning the quirks of Shotcut. So, skill issue. 😂
The video’s far from perfect, but it’s a lil keepsake in my archive that takes me back to simpler times.
Part of me loves that we have AI tools making video production so much easier. But another part of me is going to miss the masochism of obsessing over every tiny detail to bring something to life exactly how you envisioned it.
youtu.be/PgwFNMl16hY?is=7DIW…
Dario is a confused man who desperately needs to install a PR person in his head to filter how he communicates.
I understand that he genuinely believes what he says, but where he fails is reading the room. A majority of Americans already believe AI will do more harm than good.
So every time he opens his mouth, he’s just pouring rocket fuel on an already raging fire.
But maybe that’s the point. Maybe he secretly wants adoption rates to remain low, buying himself more time to build aligned ASI so he can save humanity from the very technology he’s helping create.
Obviously speculation, but boy, it sure makes you wonder. 🌚
I'm building the most intense cooking game ever created for the @joinHandshake🤝🏾@OpenAIDevs
Can't wait to share!! My dot has been cooking up some amazing 3D assets in Blender on its VM.
100 follower milestone! You know what that means?
I can officially retire now. 🥲
p.s. just noticed someone else snuck in the 101 before I could post. Need someone to unfollow😂
People are in awe of the discrepancy between Opus 5.5 and OAI models, as if Ant has some secret special sauce.
But it’s no secret. That special sauce is your data.
Claude Code had roughly a one-year head start on Codex in meaningful developer adoption.
That’s a year of accumulating real-world SWE data, engineering feedback, corrections, and workflow patterns that could help improve subsequent models.
The end result is what you’re seeing now with the quality of Opus’ output.
It’s also why I speculate OpenAI chose to abruptly sunset Sora and pivot so aggressively toward Codex.
They recognized that you can’t remain at the frontier if you can’t trust the code your models produce.
A model that can’t reliably engineer software can’t reliably improve itself. And without the real-world SWE data developers provide to close that gap, there’s no credible path to RSI, AGI, or ASI.
p.s. it's cheaper to provide resets than to purchase that data. 🌚
Been sitting on this idea for a while and finally decided to put a visual to it.
A rough concept for where I think agentic interfaces could be heading.
I don’t think chat is the final form of agent UI, but it’ll likely remain one medium of input because natural language is such an intuitive way to express intent.
What I think needs to evolve is how that intent gets interpreted and structured into something more like object formation.
You communicate with the system, and it creates or updates durable objects: projects, tasks, files, decisions, research, timelines, agents, and the relationships between them.
So rather than “everything is chat,” I envision something closer to:
→ Communicate intent.
→ Intent creates or evolves existing objects over time.
→ Those objects preserve context and continuity, giving agents something persistent to reason over and execute from.
I mocked up a UI concept called Cairn that feels closer to this direction than a pure chat interface.
Note: The prototype is considerably more bloated than what I actually envision. I’m just too lazy to refine it right now 😂
