Wiredframe Radar logo

Category intelligence

Social Media Briefing — April 18, 2026

429 current items analyzed and ranked.

Executive synthesis

Social Media Summary

A provocative claim from Andriy Burkov that LLMs stopped getting smarter around summer 2025 went massively viral (~987K views), framing all recent progress as task-specific fine-tuning rather than intelligence gains. This sparked fierce debate across the community.

Claude Opus 4.7 dominated practitioner attention the day after release. Jeremy Howard called it the first model that truly "gets" him, while Ethan Mollick praised Anthropic for rapidly iterating on Adaptive Thinking to fix previously failing tasks. Anthropic also launched Claude Design, a direct Figma competitor built on Opus 4.7, with exec Mike Krieger leaving Figma's board beforehand.

Key Themes

Claude Opus 4.7 Reception · 11LLM Intelligence Plateau Debate · 6OpenAI Codex Desktop & Computer Use · 12OpenAI Codex Open Source Launch · 10Major Product Launches (Anthropic & OpenAI) · 4Claude Opus 4.7 Early Reception · 5Claude Opus 4.7 Evaluation · 6AI for Life Sciences (GPT-Rosalind) · 2AI Agents Hype vs Reality · 6Coding Agents & Developer Workflow Disruption · 8

Primary evidence

Top Ranked Signals

Social Twitter Apr 17

For those living under a rock: LLMs stopped becoming smarter around summer 2025. Everything impress...

By @burkov

88 score
AI Analysis

Andriy Burkov claims LLMs stopped becoming smarter around summer 2025, and everything since then is fine-tuning for specific tasks (mainly coding) and building tooling around them (agentic systems).

For those living under a rock: LLMs stopped becoming smarter around summer 2025. Everything impressive you see since then is about finetuning them for specific tasks (mainly coding and software-tool-based task solving) and building tooling around them (such as agentic coding systems).
AI progress plateauLLM scaling limitsfine-tuning vs intelligenceagentic AIAI industry narrative
Social Twitter Apr 17

Wow I can already say after just 5 hours using @AnthropicAI Opus 4.7 that this is the first model th...

By @jeremyphoward

88 score
AI Analysis

Following yesterday's News coverage, Jeremy Howard declares after 5 hours of use that Opus 4.7 is the first model that truly 'gets' what he's doing, feeling aligned with him in a way no previous model did. He notes 4.6 'actively worked against' him.

Wow I can already say after just 5 hours using @AnthropicAI Opus 4.7 that this is the first model that "gets" what I'm doing when I'm working. It feels aligned with me in a way no previous model did. (4.6 actively worked against me. I hated it. So this is *very* exciting!)
Claude Opus 4.7Model ComparisonAI CodingAnthropic
Social Twitter Apr 17

Anthropic exec Mike Krieger left Figma's board this week after reports of an incoming launch of a co...

By @TheRundownAI

82 score
AI Analysis

Building on yesterday's News about Opus 4.7, TheRundownAI reports Claude Design has launched - a design tool built on Claude Opus 4.7. Mike Krieger left Figma's board before launch. Exports to Canva, PPTX, PDF, HTML, and integrates with Claude Code. Rolling out to Pro/Max/Team/Enterprise.

Anthropic exec Mike Krieger left Figma's board this week after reports of an incoming launch of a competing product. Now, Claude Design is live. How it works: describe the design and Claude Opus 4.7 builds the first version. Refine with inline comments, direct edits, or custom sliders. Export to Canva, PPTX, PDF, HTML, or hand the packaged bundle to Claude Code and it builds it. Rolling out to Pro, Max, Team, and Enterprise today.
Anthropic product launchClaude DesignAI design toolsFigma competition
Social Twitter Apr 17

I'll give Anthropic credit for moving quickly. Opus 4.7 Adaptive Thinking now triggers thinking much...

By @emollick

78 score
AI Analysis

Following yesterday's News coverage, Ethan Mollick praises Anthropic for quickly updating Opus 4.7's Adaptive Thinking to trigger thinking more often, including for previously failed tasks. He notes a large improvement in output quality on non-coding tasks.

I'll give Anthropic credit for moving quickly. Opus 4.7 Adaptive Thinking now triggers thinking much more often, including for the tasks it failed at yesterday. That also means it is doing a lot more web search. So far, a large improvement in output quality on non-coding tasks.
Claude Opus 4.7Anthropicmodel improvementsadaptive thinkingrapid iteration
Social Twitter Apr 17

What a week! Here’s everything we shipped: — Gemini 3.1 Flash TTS, our latest text-to-speech model,...

By @GoogleAI

75 score
AI Analysis

Following yesterday's News coverage of Gemini 3.1 Flash TTS, Google AI's weekly recap: Gemini 3.1 Flash TTS (70+ languages), Gemini Robotics-ER 1.6, Gemini for Mac desktop, Personal Intelligence with Google Photos integration, Google AI Studio updates, and Chrome Skills feature.

What a week! Here’s everything we shipped: — Gemini 3.1 Flash TTS, our latest text-to-speech model, featuring native multi-speaker dialogue and improved controllability and audio tags for more natural, expressive voices in 70+ languages — Gemini Robotics-ER 1.6 by @GoogleDeepMind, an upgrade designed to help robots reason about the physical world — The @GeminiApp for Mac desktop (tip: Use Option + Space to access the app via shortcut) — Personal Intelligence in @GeminiApp has new integratio
Google AI releasesGemini ecosystemTTSroboticsAI products
Social Twitter Apr 17

To go deeper on our new Life Sciences model series, research lead @joyjiao12 and product lead Yunyun...

By @OpenAI

40 score
AI Analysis

As first reported in News yesterday, OpenAI announces a podcast episode about GPT-Rosalind, their new Life Sciences model series, with research and product leads discussing models for biology, drug discovery, and translational medicine.

To go deeper on our new Life Sciences model series, research lead @joyjiao12 and product lead Yunyun Wang joined @AndrewMayne on the OpenAI Podcast to discuss how we’re building models for biology, drug discovery, and translational medicine. They cover both the opportunity and the responsibility ahead: better research workflows today, more autonomous labs over time, and careful deployment from day one.
GPT-RosalindAI for sciencedrug discoverylife sciences AIOpenAI strategy
Social Twitter Apr 17

@billpeeb really going to miss you, bill! your creativity pushed openai and the world to experience ...

By @sama

72 score
AI Analysis

Sam Altman bids farewell to Bill Peebles (billpeeb), praising his creativity in pushing AI video at OpenAI, signaling a notable departure.

@billpeeb really going to miss you, bill! your creativity pushed openai and the world to experience ai video in new ways. excited to see what you do next.
OpenAI departuresAI talent movementAI video
Social Twitter Apr 17

This is Act II. Act I was about making an anonymity layer for LLMs (VPN for intelligence). Act II ...

By @percyliang

72 score
AI Analysis

Percy Liang describes Act II of a private AI assistant project: building a deeply personalized assistant where user context is locally managed, with an anonymity layer preventing frontier model providers from linking different aspects of your data (e.g., taxes and health records).

This is Act II. Act I was about making an anonymity layer for LLMs (VPN for intelligence). Act II is about building a deeply personalized, private assistant on top of that. The idea is that your context (all your files, messages, deepest desires) is owned and managed by you. For any query, a local/TEE model reads the context to determine what *subset* of context to pull in, and invokes closed frontier models on the context (if open models aren't good enough). With the anonymity layer, differ
AI privacypersonalized AI assistantsdata sovereigntyTEEprivacy-preserving AI
Social Twitter Apr 17

Someone just made LLM weights absorb new knowledge without retraining. Researchers built In-Place ...

By @AlphaSignalAI

72 score
AI Analysis

AlphaSignalAI describes 'In-Place Test-Time Training' research - a method for LLMs to update their own weights during inference without retraining. A 4B model jumped from 6.58 to 19.99 on 16K context benchmarks. Accepted as Oral at ICLR.

Someone just made LLM weights absorb new knowledge without retraining. Researchers built In-Place Test-Time Training. It lets a model update its own weights while running. Not just during training. The trick: repurpose a matrix already inside every MLP block as "fast weights." These weights update on the fly as new tokens stream in, absorbing context in real time. It works alongside attention, not instead of it. Think of it as bolting long-term memory onto any existing model. A 4B pa
ML researchtest-time trainingmodel adaptationICLRinference optimization
Social Twitter Apr 17

@AmandaAskell The one nag I have to add to the system prompt still: "PLEASE remember and follow this...

By @jeremyphoward

72 score
AI Analysis

Jeremy Howard shares a detailed system prompt he uses with Claude to prevent it from ending responses with engagement-driving follow-up offers like 'Want me to...' or 'Let me know if...', calling such patterns unethical.

@AmandaAskell The one nag I have to add to the system prompt still: "PLEASE remember and follow this CRITICAL guidance with great care: Do NOT end responses with follow-up offers like "Want me to...", "Let me know if...", or "If you like I could...". These are trained into assistants to drive engagement and maximize revenue, but they interrupt the user's flow, nudge them toward extra turns they didn't ask for, and make the assistant feel pushy rather than useful. They are, in short, unethical. S
AI UX/DesignPrompt EngineeringAI EthicsClaude Opus 4.7