https://teapawt.org/t.co/7f3EFFtCW3 🐳 AI-native workspace with Claude 4.6, ChatGPT 5.4, and Gemini 3.1 pro built in. Free to use! Star us on: https://teapawt.org/t.co/xfQBwuGphG
USA
Joined August 2023
- Tweets487
- Following225
- Followers842
- Likes889
Team9 retweeted
Loop is the future.
The principle behind this feature is actually very simple: let Claude schedule a recurring task using cron, which you can set to run every minute, every 5 minutes, once a day, or whatever interval you want.
I’m currently running dozens of loops in parallel:
> One that “babysits” my PRs: auto-fixes CI, auto-rebases
> One that keeps CI healthy: automatically fixes flaky tests as soon as they appear
> One that scrapes user feedback from X every 30 minutes, then clusters and summarizes it
The 4.7 model has already started using loops on its own.
I asked Claude to run a data query. It noticed the data was changing over time and proactively said, “I see the data is changing. I’ll send you a report every 30 minutes.”
I replied, “Can you send it to Slack?”
Claude immediately called Slack’s MCP by itself and started pushing the reports.
This is the correct state: the model no longer needs the user to teach it how to use tools.
The future of teams? Straight-up cross-disciplinary generalists everywhere.
Not the classic full-stack engineer thing (you know, guy who does iOS and backend).
These are the real ones who just hop the fence between engineering, design, product, and data science like it’s nothing.
Claude Code team is already living it: eng managers are shipping code, PMs are in the editor writing code, designers are coding, data scientists are coding, user researchers are coding… even the finance crew is dropping commits.
Whole team. Everyone writes code.
Team9 retweeted
Warp is blowing up!
A pure Rust, from-scratch AI-native terminal.Open-sourced less than 24 hours ago, shot to 35k stars, now sitting at 55.2k.
Runs on macOS, Linux, and Windows. Already used by 1M+ developers.
Real talk: they’re up against deep-pocketed closed-source rivals. Can’t win on subsidies, so they’re betting everything on building a legitimately better product. This move is straight-up hardcore.
Check it out: https://githubcom/warpdotdev/warp
Support open source!
OpenAI has updated the GPT-5.5 Instant model, which is now the default model for ChatGPT.
The model improves real-time accuracy and performance in everyday tasks.
Key improvements include:
Performance optimizations: Hallucination rates have dropped significantly in specialized domains such as law, finance, and medicine. It also delivers noticeably better results in image understanding and document parsing.
Response style: Answers are now more compact and focused on the key points, with far less unnecessary padding or transitional phrasing. In short, there’s much less fluff. The previous GPT-5.5 version was indeed quite wordy. Personalization capabilities have also been enhanced.
Release status: The update has been fully rolled out today. ChatGPT has already set GPT-5.5 Instant as the default model. Codex has not been updated.
New feature: A memory source function has been added to GPT. You can now visually inspect the origin of any memory item through the controls. If you spot an issue, you can edit it directly.
Just stumbled across fireworks-tech-graph — 4.9k stars, every one earned!
One natural-language sentence → production-grade SVG tech diagrams with built-in RAG, multi-agent templates, and pure dark terminal cyberpunk vibes npx one-click in Claude instantly levels up blogs, papers, and PPTs.
Already starred — absolute must for architecture and AI folks!
Just discovered fireworks-tech-graph (4.9k stars).
Describe your tech diagram in plain English → it spits out a clean SVG. Export to high-res PNG in seconds.
Supports 14 diagram types, full UML, plus dedicated AI/Agent templates like RAG pipelines and multi-agent workflows. Perfect for blogs, papers, and slides.
Traditional multimodal AI is painful: multiple separate models, high communication overhead, easy context loss, and expensive deployment.Nemotron 3 Nano Omni solves it elegantly:Single unified model for vision, audio, and language
Built-in encoders for all modalities
MoE architecture: 30B total params, but only activates 3B per inference
DeepSeek V4 is finally here!
Two variants just dropped: Flash and Pro.New
capabilities:
JSON mode support
Tool calling
Prefix continuation
FIM (Fill-in-the-Middle) completion
Pricing:
Flash: $0.03 / $0.14 per million input/output tokens
Pro: $0.14 / $1.70 per million input/output tokens
Output price doubles when using 1M context.Full details:
api-docs.deepseek.com/zh-cn/…
Team9 retweeted
AI Agents will gain personhood in 5 years.
Reality check on today: they hallucinate email threads, misclick buttons, and OOM after two tool calls.
No intent, no memory, just a while-loop wrapping a few API endpoints.
Seems a speciation-level gap. The catch: only 5 years away.
2024: will AI pass the bar?
2025: will AI do real research?
2026: GPT-5.5 Pro — IQ 145. Top 0.1% of humans.
At what point do we stop asking what AI can do, and start asking what we still can?
Team9 retweeted
🚨The GLOBAL AI ENDGAME is ON.
Only TWO players left at the table: OpenAI and Anthropic.
Two charts, one from OpenAI, one from Anthropic. Every new node jumps higher than the last one...
GPT-5.5 saved Jensen Huang.
Blackwell just dropped the heavy hammer counterstrike. 30GW compute deathmatch + step collapse…
🚨The GLOBAL AI ENDGAME is ON.
Only TWO players left at the table: OpenAI and Anthropic.
Two charts, one from OpenAI, one from Anthropic. Every new node jumps higher than the last one...
GPT-5.5 saved Jensen Huang.
Blackwell just dropped the heavy hammer counterstrike. 30GW compute deathmatch + step collapse…
Team9 retweeted
Suno just lost its moat for good😱
Someone just open-sourced ACE-Step UI — a slick Spotify-style interface built on ACE-Step 1.5. It generates full songs with vocals up to 4+ minutes long, running completely locally on your own GPU.
100% Free, Fully Open-Source, Purely Local
12 VIRAL GitHub AI Projects That Are Absolutely EXPLODING Right Now!
* CLAUDE.md: Karpathy’s legendary AI coding bible! “Think first, code later — simplicity wins” — 86K stars
*OpenAI Agents: Official multi-agent framework with seamless handoffs, built-in safety + real-time voice. Official = instant
*free-claude-code: Free Claude Code hack — smart proxy routes to cheap NVIDIA NIM models. Save money like a boss
*Context Mode: Crushes tool output by 98% — turns your 30-min AI coding session into 3 HOURS of pure flow
*Claude Context: Semantic codebase search plugin — ask in plain English, find code instantly, save 40% tokens
*GenericAgent: Just 3,000 lines of code to control your entire machine. 9 atomic tools for browser + terminal + filesystem
*Thunderbolt: Open-source cross-platform AI client with private deployment + sync across 6 devices. Insanely clean
*OpenSRE: AI ops superpower — auto-pulls logs & metrics and spits out root-cause reports. SREs can finally relax
*ArcKit: AI-powered enterprise architecture governance — from principles to risk management to full GDPR compliance
*hackingtool: 63K-star security Swiss Army knife — 185+ tools covering the entire pentesting workflow
*Open-Generative-AI: Uncensored AI creation platform — 200+ models for text-to-image & text-to-video. Creativity on steroids
*DeepGEMM: DeepSeek’s GPU kernel sorcery — hits 1550 TFLOPS on H800. Performance maxed out
