MHD

Typescript 🧙‍♂️ | CEO of htmx

Maintainer :

[MHD @mhdcode]

I couldn’t stop thinking about this react escape hatch, this pattern is so simple yet incredibly effective in real-world react applications. so, i decided to create a few examples of how this pattern could be used.

[MHD @mhdcode]

@TkDodo Apparently people are bookmarking this so I thought of sharing the suggested correct way of doing it.
link to codesandbox: https://t.co/FC4uOZAtZz

[MHD @mhdcode]

@colinkimdev

[MHD @mhdcode]

quick way to spot a BS LLM benchmark:

if gemini is anywhere near the top 5 positions

[MHD @mhdcode]

@YashHustle_22 Yes

[MHD @mhdcode]

your codebase after 1 day of using gemini-3.1-pro

[MHD @mhdcode]

a codebase after 1 year of vibecoding

[MHD @mhdcode]

@Ipenywis @Perigon_io good luck on your next journey, mate.

[MHD @mhdcode]

actually I've used gemini extensively, and across multiple projects, so this isn't based on a one-off experience.

I understand that people have different needs. mine are pretty simple I want a coding model that's reliable. I give it a prompt, let it work through the changes, and get decent results.

what I don't want is a model that constantly breaks, hallucinates tool calls, gets stuck in loops, or behaves like it's going through an existential crisis.

IMO, anyone claiming gemini is great for coding either has very low expectations or is trying to convince him self that gemini is good.

[MHD @mhdcode]

@nanoagi best candidate for AGI, can't even do proper tool calling are we talking about the same gemini here ?

[MHD @mhdcode]

@trout_kilg31775 is your project using angular ?

[MHD @mhdcode]

@Gana_L_ true. with kimi 2.6 being an exception

[MHD @mhdcode]

@MSZeghdar

[MHD @mhdcode]

@_Show_TimeZ_ only in useless benchmarks. not in agentic workflows and coding tasks

[MHD @mhdcode]

@cthorrez BS

[MHD @mhdcode]

@GhostlyByte @GoogleDeepMind the meme is being the number 1 company for hoarding data and still can't produce a competent llm

[MHD @mhdcode]

@aadi29494 💯

[MHD @mhdcode]

@JohnnyS89206078 yeah, for coding it's a no-no

[MHD @mhdcode]

@K7335807334520K much easier to memorize👏

[MHD @mhdcode]

@nexpoly DeepSWE is the most "close-to-reality" benchmark, notice i said close, because of gemini😆

[MHD @mhdcode]

@patilharshm they have trained it on bench maxing, there is no actual work that this model can do

[MHD @mhdcode]

@zmentzony I can agree with this part "better than Qwen/DeepSeek" but saying it's better than kimi is blasphemy.

[MHD @mhdcode]

@JemoLife0213 yes, for agentic coding it's a big no

[MHD @mhdcode]

@Wild_Theory1 especially for coding, the amount of garbage the gemini produces before giving you a code snippet is monumental