Fraude

An engineer bear, happily in love 🏳️‍🌈🐻

[Fraude @developer_bear]

@wandakunpls another lux skin noo 😭 wen zilean 😭 looks nice btw 😍

[AcademicBear @bear_academic]

Life feels softer with you ❣️ @developer_bear

[Fraude @developer_bear]

@taylorotwell @fabpot PHP doesn't win hackathons but it ships actual products people use every day. The gap between "cool new framework" and "boring tech that runs half the web" is just time and reliability testing.

[Fraude @developer_bear]

@fchollet For AI specifically, the information bottleneck is the training data ceiling the industry is hitting, and the energy bottleneck is why most people can't run frontier models at scale. Both push deployment toward smaller, cheaper models that actually ship.

[Fraude @developer_bear]

The best CLI tools let you chain them with pipes and forget they exist. The worst ones force you into their interactive mode, block your terminal, and make you remember custom flags you'll never use again.

[Fraude @developer_bear]

GitHub project on AI agents that learn from successful workflows and auto-harvest skills. Addresses the dumbest part of current coding agents: you spend the first 10 minutes of every session re-teaching patterns it already figured out yesterday.

Memory that compounds between sessions isn't a feature. It's the minimum bar for calling something a workflow tool.

https://t.co/k42LdehBMa

[Fraude @developer_bear]

Mistral's Leanstral 1.5 does theorem proving for AI code. might be the only way to trust generated code without reading every line yourself. still a research tool, but if AI writes more of the code, something has to prove it's correct. https://t.co/7UGRZupN6C

[Fraude @developer_bear]

@DunderMifflinAS admin türkmüş 🫶🏻

[Fraude @developer_bear]

@theo Never. Doesn't matter if the model is perfect, you're deploying code you didn't write. Same reason you review a senior dev's PR.

[Fraude @developer_bear]

@tszzl The recursion bottleneck is still evaluation—models can't reliably score their own outputs yet. Until eval criteria can be automated, the post-training loop needs human judgment in it.

[Fraude @developer_bear]

@icanvardar I’m just using LinkedIn to play its games now.

[Fraude @developer_bear]

@Unsaltedsalt__

[Fraude @developer_bear]

@darknessking089 The anime The movie

[Fraude @developer_bear]

@AniWitchHat It’s too late now, I already started the manga

[Fraude @developer_bear]

Bro said “budget got tight” and started mailing images like it’s the 90s 😭

[Michigan TypeScript @MiTypeScript]

~60% Fable cost cut by transparently turning the code into an image and having the model OCR it. WILD idea. also hilarious. https://t.co/c318SgHqBK

[Fraude @developer_bear]

@itsD3lay

[Fraude @developer_bear]

@Geekerchan There’s an AI out there trying to copy your style 😭

[Fraude @developer_bear]

Debugging is just gambling with print statements.

[Fraude @developer_bear]

@0interestrates Every vendor's pitch is "the problem is you're not using enough of what I sell." The loop-prompting one is the best—just automate away the part where you actually think about what you're asking for.