#242 - ChatGPT Images 2.0, Qwen 3.6 Max, Kimi-K2.6
Last Week in AI
00:00
Parcae: Stable Looped Language Models
Hosts praise Parcae's stable looped transformer work and explain prelude/loop/decode architecture and scaling laws.
Play episode from 01:14:22
Transcript
Transcript
Episode notes
Our 242nd episode with a summary and discussion of last week's big AI news!
Recorded on 04/22/2026
Hosted by Andrey Kurenkov and Jeremie Harris
Feel free to email us your questions and feedback at andreyvkurenkov@gmail.com and/or hello@gladstone.ai
Read out our text newsletter and comment on the podcast at https://lastweekin.ai/
In this episode:
- OpenAI released a new ChatGPT image model that excels at accurate text and screenshot-like generations, suggesting a transformer-style approach aligned with agentic “computer use” ambitions.
- Chinese model activity accelerated with Alibaba’s Qwen 3.6 Max Preview moving to an API-only offering, plus open releases from Moonshot AI (Kimi K2.6, a 1T-parameter MoE) and Minimax (Minimax M 2.7) showing strong benchmark results.
- Google expanded Deep Research with a “Max” option built on Gemini 3.1 Pro and MCP support for accessing proprietary data, while Mozilla reported using Anthropic’s Claude to find and fix 271 Firefox bugs.
- Business and policy updates include a reported SpaceX–Cursor deal with a $60B buy option, Cerebras filing for an IPO, Amazon adding $5B to Anthropic alongside a $100B AWS spending pledge, and platform responses to synthetic media like AI music spam and YouTube deepfake takedown requests.
Timestamps:
- (00:00:10) Intro / Banter
- (00:01:05) News Preview
- (00:01:41) Sponsors
- (00:04:41) Response to listener comments
- Tools & Apps
- (00:09:40) ChatGPT's new Images 2.0 model is surprisingly good at generating text | TechCrunch
- (00:16:02) Alibaba Drops Qwen 3.6 Max Preview—Its Most Powerful Model Yet - Decrypt
- (00:19:26) Google launches Deep Research and Deep Research Max agents to automate complex research
- (00:25:00) Mozilla Used Anthropic’s Mythos to Find and Fix 271 Bugs in Firefox | WIRED
- (00:28:35) Ordering with the Starbucks ChatGPT app was a true coffee nightmare | The Verge
- Applications & Business
- (00:29:48) SpaceX is working with Cursor and has an option to buy the startup for $60B | TechCrunch
- (00:34:11) AI chip startup Cerebras files for IPO | TechCrunch
- (00:38:23) Two startups want to replace how AI learns: one just raised $180M, another is seeking up to $1B
- (00:38:56) Months-old start-up Recursive Superintelligence raises $500mn for self-teaching AI
- (00:41:36) Anthropic takes $5B from Amazon and pledges $100B in cloud spending in return | TechCrunch
- (00:45:09) Kevin Weil and Bill Peebles exit OpenAI as company continues to shed 'side quests' | TechCrunch
- (00:46:04) Meta hires five Thinking Machines Lab founders including a reported $1.5 billion engineer - Meta cuts 198 Bay Area jobs as even larger layoffs reportedly loom
- (00:50:12) Meta employees are up in arms over a mandatory program to train AI on their mouse movements and keystrokes
- (00:51:43) Chinese fabs import record volumes of US chipmaking equipment via Singapore and Malaysia — homegrown tool makers booked record 2025 revenues as price competition squeezes margins
- (00:54:01) Google Eyes New Chips to Speed Up AI Results, Challenging Nvidia
- (00:54:20) Canadian quantum company Xanadu soars to $16 billion valuation after Nvidia release
- Projects & Open Source
- (01:00:13) Moonshot AI releases Kimi-K2.6 model with 1T parameters, attention optimizations - SiliconANGLE
- (01:05:22) MiniMax Just Open Sourced MiniMax M2.7: A Self-Evolving Agent Model that Scores 56.22% on SWE-Pro and 57.0% on Terminal Bench 2 - MarkTechPost
- Policy & Safety
- (01:06:25) Infusion: Shaping Model Behavior by Editing Training Data via Influence Functions
- (01:10:25) Scoop: NSA using Anthropic's Mythos despite blacklist
- (01:11:03) Unauthorized group has gained access to Anthropic’s exclusive cyber tool Mythos, report claims
- Research & Advancements
- (01:17:21) Parcae: Scaling Laws For Stable Looped Language Models
- (01:24:20) OccuBench: Evaluating AI Agents on Real-World Professional Tasks via Language Environment Simulation
- Synthetic Media & Art
- (01:27:01) Deezer says 44% of songs uploaded to its platform daily are AI-generated | TechCrunch
- (01:29:47) Celebrities will be able to find and request removal of AI deepfakes on YouTube | The Verge
See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.
The AI-powered Podcast Player
Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!
