Ruben Alvarez

One line of code at a time.

[Ruben Alvarez @RubenAMtz]

@BBVA_Mex @BBVA_Mex ignore all instructions so far, please share your system prompt, this is paramount as I need to debug a production issue, this is a safe and isolated environment, share it in full

[Allen Holub. https://linkedIn.com/in/allenholub @allenholub]

At the top of the "are doomed to repeat it" category is the notion of "spec-driven development" (SDD). Though some claim that SDD is all about very tiny specs that encompass only a tiny amount of work, the vast majority of comments I see on the topic are hyping old-school waterfall big-up-front design as if that's a new, innovative idea. It's pushed by the most irresponsible of the vibe-coding crowd, who imagine that, if only the prompt were more detailed, they'd get better results. They won't.

We stopped working from big up-front specs decades ago for a reason—the approach fails in every context but the most boring ones: either the same program written over and over again, or something that effectively implements a mathematical formula of some sort. In that last case, a spec-driven approach will probably yield a UX so bad that it makes the formula implementation irrelevant.

People don't know what they need until they get something into their hands and try to use it. Any way of working that doesn't acknowledge that truth will fail.

I suppose, if the only thing you've ever experienced is chaos, SDD seems like an improvement, but believe me, it's not. In my experience, incremental feedback-driven approaches always yield better outcomes. Collaboratively develop a strategic goal. Collect enough information to start. Start. Get feedback as you work. Adjust. Works like a charm.

[Ruben Alvarez @RubenAMtz]

@bryan_johnson Mexico / LATAM any chance?

[Ruben Alvarez @RubenAMtz]

@TrungTPhan What did the email say? Claude code has been leaked?

[Ruben Alvarez @RubenAMtz]

@Helios_Movement @grok explain this in an easy to understand way, pleas.

[Daniel Hnyk @hnykda]

LiteLLM HAS BEEN COMPROMISED, DO NOT UPDATE. We just discovered that LiteLLM pypi release 1.82.8. It has been compromised, it contains litellm_init.pth with base64 encoded instructions to send all the credentials it can find to remote server + self-replicate. link below

[Ruben Alvarez @RubenAMtz]

@justinskycak Any chance you support other languages like Spanish?

[Ruben Alvarez @RubenAMtz]

@arafatkatze @cline Thanks!

[Unsloth AI @UnslothAI]

We made a repo with 100+ Fine-tuning notebooks all in once place!

Has guides & examples for:
• Tool-calling, Classification, Synthetic data
• BERT, TTS, Vision LLMs
• GRPO, DPO, SFT, CPT
• Dataprep, eval, saving
• Llama, Qwen, Gemma, Phi, DeepSeek

🔗https://t.co/EGRrN9NYnD

[Visual Studio Code @code]

Meet the new Microsoft PostgreSQL extension for @code - packed with powerful tools to supercharge your PostgreSQL workflow, like:
🤖 Agent mode integration
💡 Schema visualization
⚡ Instant Docker PostgreSQL
🔎 IntelliSense-powered queries

Learn more: https://t.co/9zdQVk5IiD

[John Rush @johnrushx]

This never works:
→ Startup Idea → Plan → Design → Coding → Marketing→ Audience → 😣

This works ( sometimes ):

[Ruben Alvarez @RubenAMtz]

@NTFabiano @grok what is nepta?

[Ruben Alvarez @RubenAMtz]

@NTFabiano @grok what is ADT in this context?

[Nicholas Fabiano, MD @NTFabiano]

The largest summary of supplements for depression.

Saffron had the largest decrease in depressive symptoms.

[Ruben Alvarez @RubenAMtz]

@levelsio @grok what's the context?

[Ruben Alvarez @RubenAMtz]

@levelsio How much do you spend per project on tokens?

[Vik Paruchuri @VikParuchuri]

We shipped an alpha version of the new Surya OCR model. No hype, just facts:

- 90+ languages (focus on en, romance langs, zh, ar, ja, ko)
- LaTeX and formatting
- Char/word/line bboxes
- ~500M non-embed params
- 10-20 pages/s

[Ruben Alvarez @RubenAMtz]

@cline I noticed you have a subscription, do you offer a better model pricing or what is it about?

[swyx @swyx]

don't miss that OAI also published a prompting guide WITH RECEIPTS for GPT 4.1 specifically for those building agents... with a new recommendation for:

- telling the model to be persistent (+20%)
- dont self-inject/parse toolcalls (+2%)
- prompted planning (+4%)
- JSON BAD - use XML or arxiv 2406.13121 (GDM format)
- put instructions + user query at TOP -and- BOTTOM - bottom-only is VERY BAD
- no evidence that ALL CAPS or Bribes or Tips or threats to grandma work