Meess

Building with a smaller model, tweaking prompt to better adhere to what to you want, then use an improved model. Delivers very good results for use (analysing documents and structuring them via structured output). Forces you to make a good prompt. Gemini flash-lite->flash->pro

@GoogleDeepMind Finally a proper replacement for GPT‑4.1 mini

Gemini 3.1 Flash-Lite has landed. It’s our most cost-efficient Gemini 3 series model yet, built for intelligence at scale. Here’s what’s new 🧵

@GoogleDeepMind Is it also preferred to keep the temperature at 1 as the other 3.1 models? Or for exact translation still the recommendation to set the temperature lower?

@OfficialLoganK copy deep research as markdown in gemini, when?

@ScriptedAlchemy With github copilot it can be beneficial to ask multiple questions in one go, as it doesn't work with tokens but per request (i.e per question asked, which can be a very long periode of the AI working autonomously as an agent)

Using subagents to route work through different gpt providers like openai, work, and github copilot

@stephenhaney Can we leverage it in VsCode with co-pilot?

Hello! Today we're releasing Paper Desktop Paper is now a canvas for Cursor, Claude Code, Codex. Any agent can read and write html to Paper. • push or pull from your codebase • pull real data from anywhere • less work, more design What will you ship? Sound on 🎶

@rauchg Right in time for migrating some production apps off Heroku

Python on Vercel is getting major upgrades, starting with 2x larger max bundle size. More to come.

@youyuxi @boshen_c What the return on investment of working out a good idea? As it's become so simple to just copy it now, demotivates to work hard on these good ideas

@boshen_c No! On the contrary, the cost of trying new ideas is infinitely scaling to zero. It’s up to us to slow down and search for great ideas.

AI coding is leading to a loss of innovation. Innovation requires long periods of refinement and those “shower thought” moments. Right now, it’s impossible to innovate with so many coding sessions open, where you are high as fuck with intracranial tinnitus that won’t stop.

@kentcdodds Serieus issue atm… I don’t have a wife and kids, glued to my desk

If I didn't have a family, church, and choir, I think I would probably just disappear in some basement and prompt AI unendingly. It's just that addicting. I'm so glad that's not my life.

@nickolsoncodes @databricks @DrizzleORM Let's first go V1...

Open source is getting scooped up. Anthropic buys Bun, Cloudflare buys Astro. What are the chances @databricks tries to buy @DrizzleORM? They already bought their bff Neon.

Tsgo

Seems like I'm kinda beta: - Vite 8 (beta) - Tanstack start V1 (beta) - Drizzle V1 (beta) - Gemini 3.1 (preview) Works amazing though (once set-up without errors)

@mattpocockuk Still to go: - Django 4.2.28 -> Django 5.2.11 - One large frontend package 2 major versions - Djangorestframework replacement (DRF seems explicitly not go into the async direction Django and Python are heading)

@mattpocockuk Yesterday finally upgraded (with codex 5.3 help): pipenv -> uv webpack -> vite8 Something our team estimated to would costs a dev over a month (due to all side-effects and edge case differences), and was therefore never planned. This makes the threshold to remove legacy so low

A possible future: For the last 30 years, devs have made the sensible assumption that every time a human touches a codebase under time pressure, the codebase gets less maintainable. This is called software entropy. It's only possible to escape by continual monitoring and improvement. Or by employing ONLY 10x devs (impossible). The current generation of LLMs aren't currently good enough to escape software entropy. Without careful supervision, they make codebases worse. But soon, they will be good enough that they will make the codebase better each time they touch it. At that point, we won't need human review any more. Codebases will be better as a black box. We'll review inputs and outputs. Before Opus 4.5, models produced software entropy at an alarming rate. Opus 4.5 was the first time I felt like the entropy was manageable. Soon, there might not be any entropy at all.

@brenelz @burkeholland 15 min ago I suddenly saw it when changing models, so seems to be rolling out rn

@burkeholland is 5.3 codex available in github copilot yet?

@boshen_c With Github co-pilot, doesn't matter how much tokens it burns, it is slow still. In general I use codex 5.2 (5.3 not rolled out there yet). Codex in general better, but Claude Opus does better way of displaying in chat, making todo lists for itself and starting sub agents.

Claude Opus 4.6 is really bad, I urge everyone to try Codex, buy the $20 plan first. 4.6 is really slow, burns tons of tokens, and is pretty dumb 😀

@ScriptedAlchemy @stolinski You still need to steer them, check if they do go in the direction / vibe you want. Double check that what you wanted works and if that what you wanted was actually what you wanted. This translation from my imagination still keeps me plenty busy,but everything does become easier

@stolinski Why? You have unlimited output capacity. My unfinished projects were all due to capacity. Biggest problem is finding more work for them to take on

My biggest problem with vibe coding is that I have too many different projects I'd like to make, so now instead of having 1 project I never finish, I have 10

@andrii_sherman @_juliushere @DrizzleORM Release the kraken! Curious to see a timeline for release, waiting for it to release to migrate to it

@_juliushere @DrizzleORM Check beta branch

Hey boys, is @DrizzleORM still being maintained?

After @GeminiApp not creating an image 3 times while Nano banana wat attached to the prompt, I tried to specifically name to use Nano banana... It did listen, but not what I wanted...

@theo For me 5.2 is way faster than 5.2 Codex, with often similar results

Been doing a lot of work in Codex but was trying a pass in Claude Code. Forgot how fast it was, finished in 2 minutes when I'm used to Codex taking 10+ for similar tasks. Excitedly went to go build the code and see the results. Build error. 🙃 Back to 5.2...