Sunday Rundown #139: Extra Connectors & Fire Water Burn
Live & Learn #4: Testing and ranking 5 AI text-to-visual makers.
Happy Sunday, friends!
Welcome back to the weekly AI news roundup.
In case you missed it, here’s this week’s Thursday deep dive:
If you’re consistently missing out on my emails, remember to check your “Promotions” tab and mark whytryai@substack.com as a “Safe Sender.”
🗞️ AI news
Here’s what happened in AI this week:
👩💻 AI releases
Alibaba released Qwen-Image-2.0-Pro, an image model with improved output quality, multilingual text rendering, and better instruction-following.
Anthropic news:
Claude Connectors for creative tools like Adobe Creative Cloud, Blender, and SketchUp let you use these apps directly in chat via natural language requests.
Claude Security lets Enterprise accounts scan codebases for vulnerabilities and get AI-suggested fixes. (In public beta for Enterprise customers.)
Cognition launched Devin for Terminal, a coding agent that runs locally on your computer but can hand off to a cloud agent when needed.
Google news:
Gemini now generates and exports files directly from chat, including Google Docs, Excel, Sheets, Slides, PDFs, Word documents, and more.
Google Translate now lets you practice pronunciation by speaking translations out loud and getting instant AI feedback.
Mistral news:
Medium 3.5 is an open-source reasoning and coding model that can be self-hosted on as few as four GPUs. (Try for free in Le Chat.)
Vibe can now run coding sessions in the cloud, so you can launch multiple jobs in parallel and come back to a finished task.
NVIDIA open-sourced Nemotron 3 Nano Omni, a small multimodal model that lets agents reason across video, audio, images, and text.
OpenAI added Advanced Account Security to ChatGPT and Codex, letting you lock your account with passkeys or physical security keys to protect sensitive data.
Poolside released Laguna M.1 and Laguna XS.2, agentic coding models built for long-horizon tasks (XS.2 is small enough to run on a single GPU).
Spotify launched a “Verified by Spotify” badge so you can tell at a glance whether an artist is human rather than AI.
xAI launched Grok 4.3 at a very low API price point and also released a new voice cloning suite.
🔬 AI research
Google news:
Ask YouTube is an experimental search feature that lets Premium US subscribers ask questions and get a mix of text summaries, videos, and Shorts.
Google Photos Wardrobe will soon be able to scan your photo library to create a “digital closet” so you can mix-and-match and virtually try on outfits.
Microsoft previewed a Legal Agent in Word that helps with contract review and redlining. (Available via the Frontier early access program.)
📖 AI resources
“Claude for Personal Guidance” [STUDY]: Anthropic’s look at how people actually use Claude along with observed sycophancy patterns.
🔀 AI random
OpenAI traced GPT-5‘s curious obsession with “goblin” metaphors to reward training for the “Nerdy” personality mode.
🤦♂️ AI fail of the week
I asked DALL-E 2 to visualize a living room scene a few seconds before the smoke alarm goes off. It sure delivered.
📹 Live & Learn #4: AI text-to-visual tools
Live & Learn is a relatively new Substack Live concept where I test and rate AI tools.
Here’s how it works:
I pick 3-5 tools in the same category
I test them live and rate them on 3-5 specific dimensions
I end up with a leaderboard based on the ratings from the session
This week, I looked at five text-to-visuals tools:
Heads up: There won’t be a Live & Learn next week.
That’s because I’ll instead be doing my Cozora expert session about practical applications of image-to-video tools on Thursday, May 7, 11 AM ET / 5 PM CET.
Cozora Discounts for Why Try AI subscribers
Free subscribers can claim 10% off Cozora here.
Paid subscribers get a 50% discount upon upgrading (worth $360/year).
I expect to return to Live & Learn the week after, so stay tuned!