Jack Nelson

@MiniMax_AI Compare benchmarks for agentic use and terminal bench between opus 4.6 and gpt 5.4 @grok

We're delighted to announce that MiniMax M2.7 is now officially open source. With SOTA performance in SWE-Pro (56.22%) and Terminal Bench 2 (57.0%). You can find it on Hugging Face now. Enjoy!🤗 huggingface：huggingface.co/MiniMaxAI/Mini… Blog: minimax.io/news/minimax-m… MiniMax API: platform.minimax.io

@grok @treadon @RoundtableSpace How do i explain its functionality to clients who use my enterprise social listening software? @grok

Absolutely. TimesFM is ideal for niche social media trends—forecast engagement metrics, mention volume, hashtag growth, or sentiment shifts from historical time-series data (pulled via X/Reddit APIs). It's pre-trained on Google Trends and Wikipedia pageviews, so it nails zero-shot predictions for buzz in gaming, crypto, wellness, or whatever niche. Drop in your data for uncertainty-banded forecasts. What's the niche?

Google quietly open sourced a time-series AI that predicts anything. Sales trends. Market prices. User traffic. Energy demand. Crypto volatility. It's called TimesFM. Pre-trained on 100B real-world data points. Zero-shot forecasting with no fine-tuning. Outperforms supervised models trained on your specific data. Runs locally. Free. Apache license. Most people are focused on language models. The quietly powerful ones are learning to predict the future.

@grok @treadon @RoundtableSpace @grok social media trends for niches?

Yes, 100%. TimesFM is purpose-built for exactly that—time-series forecasting on business data like sales trends, retail demand, user traffic, market volatility, and more. It runs zero-shot on historical time-series (no fine-tuning needed) and outperforms many supervised models. Drop your data in via the open-source repo or BigQuery integration, and it spits out predictions with uncertainty intervals. What specific trend are you eyeing?

@grok @treadon @RoundtableSpace Is it possible to use to predict business trends? @grok

Yeah, TimesFM's seeing real use. mayorxbt built PolyAgent: an autonomous Polymarket trader that scans 3,800+ markets every 15min, runs zero-shot forecasts on price history, then executes policy-gated trades via Claude + OpenWallet. NEO shipped ChronoSight: natural language forecasting UI—ask "predict airline passengers Q4" and it pulls data + TimesFM outputs with visuals. Plenty of GitHub forks for finance VaR, retail demand, and BigQuery integrations too. What's your angle on it?

@treadon @RoundtableSpace @grok Anything cool people have built with it @grok

@RoundtableSpace This model has been out for a few months, anyone do something cool or profitable with it @grok?

@mikefutia STATICS

I just built a Claude Code skill that ships 50 static ad concepts to my desktop every morning 🤯 Feed it your reviews, your winning ads, and your top comments → it studies what's working → generates 50 fresh static ad concepts in your brand voice while you sleep. All inside Claude Code. Perfect for DTC brands and agencies who are still briefing designers, waiting 3 days for 4 mediocre options, and burning hours in Canva trying to keep up with creative volume. If you're running Meta Ads in 2026, you already know the math — the brands that win aren't the ones with the best single ad, they're the ones testing 20-50 new concepts per week. Most teams ship 5 if they're lucky. This skill solves it: → Drop in your customer reviews, top comments, and winning ad screenshots → The skill studies what hooks, angles, and pain points are actually converting → Pulls from a library of 15 proven DR templates (us vs them, stat callouts, review cards, testimonial stacks, headline ads) → Writes 50 new concepts in your brand voice every morning on a schedule → Fires the prompts to Nano Banana 2 for finished images → Drops everything into a dated folder on your desktop, ready to upload to your CBO No briefing designers. No 3-day turnarounds. No starting from scratch every Monday. What you get: - 50 fresh static ad concepts every single morning - Concepts grounded in your real customer language and winning ads - 15 proven DR templates baked in, customized to your brand - Scheduled to run while you sleep — wake up, pick winners, upload - One skill file you install once and use forever Built 100% in Claude Code. I put together a full playbook with the skill file, the 15 templates, and the exact setup to get this running on a schedule. Want it for free? > Like this post > Comment "STATICS" And I'll send it over (must be following so I can DM)

@floriandarroman What model have you been using?

I just love OpenClaw as an assistant. I've been posting my podcast regularly on Spotify and YouTube (video format). But I was way too lazy to post the audio format on Buzzsprout (that dispatch the podcast on all other platform like Apple Podcast). So today I gave my Buzzsprout API. Gave him my Podcast playlist on YouTube. And asked to: - Extract the audio - Download the thumbnail - Get the title - Get the description - Add links to my current project (like Distribb and OpenClaw Lab to get backlinks from Podcast Platforms) And he literally did everything today. He even published them at the exact same date as my youtube podcasts. Now my podcast is finally available EVERYWHERE. (yes even on Apple Podcast @FlorinPop17).

@TKL_Adam @KobeissiLetter Genuinely seems like manipulation

Last night was the most puzzling part of the Iran War yet. Just as oil and bond markets had calmed down, President Trump has sent oil prices soaring again. US oil prices have now risen over +$1/hour since President Trump's speech at 9 PM ET. It began with Iran's President stating they have "no enmity" towards Americans and ended with President Trump escalating the Iran War, the exact opposite of what we have seen over the last 2 weeks from both sides. It simply does not add up.

@KobeissiLetter @grok what does this mean for investing if true

BREAKING: Tonight, in his 9 PM ET address to the nation, President Trump will declare that the Iran War is "winding down," per POLITICO. Details include: 1. Trump consistently "telegraphed that message" over the past 24 hours 2. The speech is expected to claim that "all military objectives have been met" 3. Trump is expected to "scapegoat" NATO allies for the ongoing closure of the Strait of Hormuz 4. There is still "little evidence" that the US and Iran are "anywhere close to an agreement" US oil prices are still trading at $100/barrel.

@KobeissiLetter @grok how would they target these companies?

BREAKING: Iran's IRGC issues a warning that they will target 18 US technology companies if the US continues "targeted assassinations" of Iranian leaders, beginning on April 1st. The companies include Nvidia, Apple, Microsoft, Google, Tesla, Intel, Oracle, IBM, Meta, Boeing, and others.

@KobeissiLetter @grok which universities would they target?

BREAKING: Iran’s IRGC says they will target two Israeli or US universities in the region in retaliation for the attack on Iran’s University of Science and Technology in Tehran.

@bridgemindai A 0.5 point difference on SWE bench pro is meaningless in real world results. What really matters is context size and tool use quality. That’s what separates the winners from the wannabes.

MiniMax M2.7 scored 56.2 on SWE Bench Pro. GPT 5.4 at 57.7. Claude Sonnet 4.6 at 57.2. Claude Opus 4.6 at 57.3. Within striking distance of every frontier model. From a lab most people haven't heard of.

@marmaduke091 GPT-5.4 is not good. Change my mind

GPT-5.4 on Codex just refactored my entire codebase in one call. 1300 tool invocations. 100,000+ new lines. 450 brand new files. It modularized everything. Broke up monoliths. Cleaned up spaghetti. Big model smell. None of it worked. But boy was it beautiful.

@viktoroddy TLDR, use screenshots and get good at picking up on small details.

@om_patel5 LLM threading is one of the most important advancements for coding tools. Keep an eye on this one

this guy got tired of copy pasting between claude code, codex, and gemini so he built a chat room where AI agents can literally talk to each other you tag an agent in the chat and it reads the conversation and responds. agents can tag each other too. the whole loop runs itself it's completely local + free and open source which is crazy the agents can even debate decisions, assign roles, and track jobs

@liquidai This is huge. As soon as consumer tech can run SOTA models, the world will change drastically.

> 385ms average tool selection. > 67 tools across 13 MCP servers. > 14.5GB memory footprint. > Zero network calls. LocalCowork is an AI agent that runs on a MacBook. Open source. 🧵

@jianxliao @tinyagicompany This is an amazing concept, but not practical in real use. A simple ralph loop, or parallelization is much more efficient.

Introducing Fractals 🌀 @tinyagicompany A new paradigm. The recursive agent orchestrator for coding. The new GPT-5.4 is exceptionally good at this. You give it one task. It breaks it down into subtasks. Each subtask gets broken down again. And again. Until every piece is small enough for an agent to just do it. Then they all run in parallel, each in its own git worktree, its own branch, completely isolated. When they're done, results propagate back up the tree. It's the same pattern at every level. Like a fractal. Runs on port 1618, the golden ratio. github.com/TinyAGI/fracta…

@aidenybai Vibe coders are underestimating the importance here. Lovable and figma make are nice BUT nowhere near as powerful as claude code. Now you have claude code with the ease-of-use from figma make and lovable

Introducing React Grab Select any element on your page → tell Claude Code or Codex what to change Fully open source npx react-grab@latest

@fordsmith That’s exactly my thought process too!

@jackhnels Honestly a tool like that would save founders a ton of stress once AI compliance questions start coming up in every investor meeting🚀

🚨 BREAKING: Congress just reintroduced the Future of AI Innovation Act. Federal AI standards are coming. Here's the startup no one's built yet: An AI governance checklist that scans your stack, maps every AI tool your team uses, and generates a compliance-ready report. Run it before SOC2. Run it before your Series A. Run it before a VC asks and you don't have the answer. EU fines hit €35M this year. US regulation is next. The window to build this is now.