Creative & Visual Media
Discussions highlighted rapid advancements in generative video and world models, with @javilopen defending an open-source real-time playable world generator based on Alibaba’s Wan2.2 (LingBot-World), emphasizing its significance despite criticisms on FPS and playability, drawing parallels to early image gen like Disco Diffusion.
People saying things like:
– fps is so low
– the worlds aren't playable
– some things disappear as soon as you look away
– it’s pointless…are completely missing the point. Just like they did back when the critiques were 'not a good image' or 'too many fingers'.
EVERY pixel… https://t.co/YLkjji1sEY
— Javi Lopez ⛩️ (@javilopen) February 2, 2026
@fofrAI shared experiments and failure cases with Google’s Genie 3, including “expected event” behaviors like a plane not flying high and a hand filling a glass from a tap.
I find the failure cases fascinating too.
> You control a hand trying to fill up a glass of water from a pouring tap pic.twitter.com/lnHlYtXKU3
— fofr (@fofrAI) February 2, 2026
More super interesting “expected event” tests in Genie 3 https://t.co/1EpyB4vicV
— fofr (@fofrAI) February 2, 2026
@c_valenzuelab noted the “arrival” of advanced image/video generation capabilities foreseen years ago.
We have arrived. https://t.co/v2WMK0aLHp
— Cristóbal Valenzuela (@c_valenzuelab) February 2, 2026
Content creation examples included @HBCoop_’s Veo 3.1 animations of storyboards like “Nano Banana Pro” and Midjourney-to-video transitions (“Flow State,” “Gateway”).
Storyboard ideas with Nano Banana Pro animated by Veo 3.1: pic.twitter.com/ePbM6hUQdw
— Heather Cooper (@HBCoop_) February 2, 2026
Flow State pic.twitter.com/5Xze57E0wk
— Heather Cooper (@HBCoop_) February 2, 2026
Gateway
Midjourney -> Veo 3.1 pic.twitter.com/QNzIgqCuGg
— Heather Cooper (@HBCoop_) February 2, 2026
@ailker demonstrated a FAL workflow using xAI’s Grok Imagine for fun vegetable fact videos.
I’ve been seeing these kinds of videos lately. So, I built a @fal workflow using @xai Grok Imagine Image & Video so you can make them easily! 🥕🍎
Just type in any fruit or vegetable name and watch it share a fun fact in a cool video like this one! pic.twitter.com/PFvF9h14ed
— ilker (@ailker) February 2, 2026
Software Development
Simon Willison noted a company’s shift from GitHub Copilot to Cursor and then Claude Code across 600 engineers, reflecting a broader trend in AI-assisted coding tools.
Made me realize I've not used an autocomplete LLM feature in a few months now
I'm much more likely to say in chat "add a new optional atomic=True param to that function that makes it run in its own transaction" or similar
Still micromanaging the code but at a different level https://t.co/pPZNQMfxAH
— Simon Willison (@simonw) February 2, 2026
Anyone managed to get Claudeception working recently, that feature where a Claude Artifact running inside the Claude consumer apps can make calls to Anthropic's own API?
I only ever get "Invalid response format" errors back when I try it at the moment
— Simon Willison (@simonw) February 2, 2026
He described preferring conversational prompts for code edits, like adding parameters, over autocomplete features, reducing direct code micromanagement.
Made me realize I've not used an autocomplete LLM feature in a few months now
I'm much more likely to say in chat "add a new optional atomic=True param to that function that makes it run in its own transaction" or similar
Still micromanaging the code but at a different level https://t.co/pPZNQMfxAH
— Simon Willison (@simonw) February 2, 2026
Willison also reported issues with Claudeception, where Claude Artifacts in consumer apps fail to call Anthropic’s API, yielding “Invalid response format” errors.
Anyone managed to get Claudeception working recently, that feature where a Claude Artifact running inside the Claude consumer apps can make calls to Anthropic's own API?
I only ever get "Invalid response format" errors back when I try it at the moment
— Simon Willison (@simonw) February 2, 2026
@goose_oss shared tips for building MCP Apps from their experience with @MCPJams, covering theme adaptation, model context control, loading states, sync, and tool restrictions.
We teamed up with @MCPJams to share everything we learned building MCP Apps.
Here's some tips that would've saved us hours:
→ Adapt to host themes
→ Control what the model sees
→ Handle loading states properly
→ Keep the model in sync
→ Restrict who can trigger toolsLink… pic.twitter.com/7KsrG0gMFF
— goose (@goose_oss) February 2, 2026
Automation & Orchestration
@altryne spotlighted OpenAI’s new Codex macOS app as a “command center for building with agents,” offering terminal-like power without the terminal.
A start of an intense week?
OpenAI launches an app for Codex 👀
The power of terminal tools without terminal https://t.co/VSpeAtHLtN
— Alex Volkov (Thursd/AI) (@altryne) February 2, 2026
Dan Shipper promoted OpenClaw Camp, featuring @nateliason’s agent orchestration system that maintains context over time, with a live walkthrough scheduled for February 6.
“who exactly is getting boiled here?”
💀 https://t.co/sq13G7z3pL pic.twitter.com/TT3U6Qvq3X
— Dan Shipper 📧 (@danshipper) February 2, 2026
Machina (@EXM7777) highlighted the potential of agentic assistants like OpenClaw, noting their ability to learn skills, download tools, and handle tasks without requiring coding knowledge, making them suitable for daily workflows.
assistants like Openclaw will be extremely powerful in the hands of great developers… but it will be life-changing for everyone else
because it's already really good out of the box, can learn its own skills and download whatever it needs, you don't need to know ANYTHING about…
— Machina (@EXM7777) February 2, 2026
i use gpt5.2 as a main model and opus 4.5 for coding, never ran out of credits so far
— Machina (@EXM7777) February 2, 2026
They emphasized using models like GPT-5.2 for general tasks and Opus 4.5 for coding within such systems. OpenRouter (@OpenRouterAI) launched “openrouter/free,” a router that intelligently selects free LLMs supporting agentic features such as tool calling, image understanding, and structured outputs.
New router: openrouter/free
The easiest way to route to all free LLMs, selected for compatibility with your request 👇 pic.twitter.com/H7ya7IGfAb
— OpenRouter (@OpenRouterAI) February 2, 2026
Strategy & Ecosystem
Satirical takes on consolidation trends emerged, with @altryne reacting to SpaceX’s acquisition of xAI.
— Alex Volkov (Thursd/AI) (@altryne) February 2, 2026
@c_valenzuelab joked about merging RunwayML’s rocket company with its AI arm.
Excited to announce we are also planning to merge our rocket company RunwayML with our AI company Runway AI. Details to follow.
— Cristóbal Valenzuela (@c_valenzuelab) February 2, 2026
Emad Mostaque coined “promptcholy,” a term for the sadness of AI generating superior ideas, inspired by Sam Altman’s experience building an app with Codex and feeling “useless.”
𝐩𝐫𝐨𝐦𝐩𝐭𝐜𝐡𝐨𝐥𝐲 /ˈprɒm(p)̲𝚝̲.̲𝚔̲ə.̲𝚕̲̲𝚒̲/ 𝘯.
A quiet sadness upon discovering that an AI you prompted has generated ideas better than your own; the peculiar uselessness of having summoned something that renders you a little less necessary. https://t.co/Qe2Odjl1RR— Emad (@EMostaque) February 2, 2026
He elaborated it as feeling “optional” regarding future relevance rather than past impostor syndrome.
Kinda the opposite of imposters syndrome, not a doubt on the merits of past achievements but rather relevance of future ones.
Rather than feeling like a fraud one feels.. optional?
— Emad (@EMostaque) February 2, 2026
Mostaque speculated on IPO’ing SpaceX alongside Grok 5 launch to match Tesla’s market cap for a merger enabling majority control.
IPO SpaceX with Grok 5 launch => ramp the share price to a Tesla merger for effective majority control?
SpaceXai market cap should be = Tesla market cap right now…
— Emad (@EMostaque) February 2, 2026
OpenRouter (@OpenRouterAI) announced a new free router model, enabling easy access to compatible free LLMs while respecting user privacy settings, marking an emerging trend in accessible AI infrastructure.
New router: openrouter/free
The easiest way to route to all free LLMs, selected for compatibility with your request 👇 pic.twitter.com/H7ya7IGfAb
— OpenRouter (@OpenRouterAI) February 2, 2026