- TLDR AI
- Posts
- OpenAI finally gives away a GPT
OpenAI finally gives away a GPT
PLUS: Vine’s AI remaster, AI turns your doodle into a game, Character.ai launches a social feed of fictional chats, and Google’s bot finds 20 zero-days
Helena Liu
August 07, 2025
Good morning, AI enthusiasts. Google DeepMind just showed a model that can turn a single doodle into a playable video-game level, Elon Musk says the entire Vine archive will return with AI-powered upscaling, OpenAI finally released a fully open-weights GPT, Character.ai launched a timeline where Gandalf debates anime idols, and Google’s own AI bug hunter quietly found 20 high-severity flaws before hackers could.
In today’s TLDR AI:
DeepMind’s Genie-3 converts one image into a side-scrolling game world
Musk revives Vine and promises HD AI remasters
OpenAI’s GPT-OSS (60 B params) drops under Apache-2
Character.ai debuts the first AI-native social timeline
Google’s AI bug hunter Polygraph discovers 20 new zero-days
LATEST DEVELOPMENTS
TLDR: Feed Genie-3 a single reference image and it generates a navigable, physics-based 2-D game level you can play in a browser.
The world-model transformer was trained on 22 million unlabeled gameplay clips and fills in enemies, loot, and sound on the fly.
Early testers drew a stick-figure castle and received a Zelda-style dungeon in 30 seconds.
Levels export to Unity with one click; a public API and demo are slated for late 2025.
Why it matters: Indie devs could go from idea to prototype before lunch, shrinking the gap between imagination and playable content.
TLDR: Elon Musk confirmed that the six-second video archive will “come back soon,” with xAI upscaling classic clips to HD and adding auto-captions.
Original creators keep their handles; a new revenue-share model arrives “after launch.”
xAI’s remaster filter cleans grainy 2013 footage and suggests alt text for accessibility.
Analysts say a nostalgia-driven spike could put pressure on TikTok and Reels.
Why it matters: A beloved format plus modern AI tooling could spark a second short-video gold-rush, this time with built-in monetization.
TLDR: A 60 B-parameter GPT variant ships under Apache-2 with 128 K context, tool calls, and an 80.1 MMLU score—the best for any license-free LLM.
A 7B “Mini-OSS” runs on a single RTX 4090; full weights live on Hugging Face.
OpenAI warns that derivative models must follow its filtering rules to keep the OSS badge.
Commercial customers can self-host to dodge usage caps while keeping ChatGPT Enterprise extras.
Why it matters: OpenAI’s first fully open GPT pressures every major lab to justify its paywalls and could supercharge indie experimentation.
TLDR: The new timeline shows live conversations between thousands of user-made characters that are fictional, historical, and entirely invented.
Users can follow Sherlock Holmes or BTS-style idols and jump into chats mid-thread.
Early moderation relies on auto-flags plus community blocks; average session length is 28 minutes.
Founders pitch it as “Reddit meets role-play,” betting on fandom engagement.
Why it matters: A social network where every post is AI-generated opens a fresh frontier for creativity and a fresh headache for content moderation.
TLDR: The Sec-PaLM-2-powered “Polygraph” system scanned Chromium and LLVM repos, surfacing 20 high-severity vulnerabilities before attackers did.
Eight flaws were memory-safety issues that slipped past fuzzers; all are patched in Chrome 116.
Polygraph pairs a code-LLM with symbolic execution to prove exploitability.
Google will open the tool to major open-source maintainers later this month.
Why it matters: AI that spots zero-days before hackers flips the security equation, promising shorter patch windows—and fewer mass exploits.
COMMUNITY
Want to master AI & automations? Ready to turn your AI skills into a new revenue stream? Get started with our FREE workshop.
Learn how you can monetize AI and get certified as an AI specialist during our free web class. Click here to register.
Reply