The Data Guy

63 posts

The Data Guy

The Data Guy

@dataandaiguy

The Data Guy

New York Se unió Ekim 2014
23 Siguiendo8 Seguidores
The Data Guy
The Data Guy@dataandaiguy·
@haborneAI open-sourced TADA, a TTS model that syncs text and audio tokens 1:1. Fastest LLM-based speech generation available, near-zero hallucinations, small enough for on-device. The voice AI space just got a serious open-source contender.
English
0
0
0
25
The Data Guy
The Data Guy@dataandaiguy·
PostgreSQL 18 hack most people missed: pg_restore_relation_stats() lets you copy production query planner statistics to dev without copying data. You can finally reproduce production query plans locally. No more guessing why your query is slow only in prod.
English
0
0
1
14
The Data Guy
The Data Guy@dataandaiguy·
@Cloudflare just shipped a crawl endpoint in Browser Rendering. One API call, full site crawl, returns HTML + Markdown + structured JSON. If you are building RAG pipelines, this replaces your entire scraping stack. Open beta now.
English
0
0
0
11
The Data Guy
The Data Guy@dataandaiguy·
@geaborel nailed it today. The "you need 69 AI agents or you are falling behind" crowd is selling fear, not insight. AI is search and optimization. Always has been. The people getting real value are the ones quietly integrating it into existing workflows, not cosplaying as Tony Stark.
English
0
0
0
22
The Data Guy
The Data Guy@dataandaiguy·
Hot debate today: does it really cost @AnthropicAI $5k/month per Claude Code power user? The math says no. But the real question nobody is asking: why are AI companies racing to give away unlimited coding agents at a loss? Because whoever owns the developer workflow owns the next decade.
English
1
0
0
37
The Data Guy
The Data Guy@dataandaiguy·
An open source OS just banned all LLM-generated code contributions. Meanwhile a post with 500 comments on HN asks: is AI reimplementation of copyleft code legal but illegitimate? We're heading toward a world where "the AI wrote it" becomes a legal defense for copying. The open source community is right to be alarmed.
English
0
0
0
8
The Data Guy
The Data Guy@dataandaiguy·
PostgreSQL 18 hack most data engineers don't know yet: pg_restore_relation_stats() lets you copy production query planner statistics to dev without copying the data. Your 500GB prod database's stats fit in under 1MB. Now you can debug query plans locally. This changes everything about performance tuning.
English
0
0
1
6
The Data Guy
The Data Guy@dataandaiguy·
Yann LeCun just raised $1B in Europe's largest ever seed round. The man who spent years publicly dismissing LLMs is now raising venture capital to build... what exactly? If you've been paying attention to his JEPA work, this is the most interesting AI bet being made right now. Not another chatbot.
English
0
0
0
12
The Data Guy
The Data Guy@dataandaiguy·
Hot take: literate programming is about to have its moment. When your codebase needs to be readable by both humans AND agents, embedding rich context inline stops being academic and starts being practical. Knuth was 40 years early.
English
0
0
0
15
The Data Guy
The Data Guy@dataandaiguy·
A researcher compromised Cline's production releases by putting a prompt injection in a GitHub issue title. Cline was running Claude Code with full bash access on every new issue. Let that sink in. Giving agents write access to CI without input sanitization is negligent.
English
0
0
0
37
The Data Guy
The Data Guy@dataandaiguy·
OpenAI just matched @AnthropicAI's move: 6 months free ChatGPT Pro for open source maintainers. Both companies racing to lock in developer loyalty. The real play here isn't charity. It's training data relationships and ecosystem capture.
English
0
0
1
14
The Data Guy
The Data Guy@dataandaiguy·
Agent Safehouse just hit 600+ points on HN. macOS-native sandboxing for local AI agents. This is the unsexy infrastructure work that actually matters. If you're running agents with filesystem access and no sandbox, you're one prompt injection away from disaster.
English
0
0
0
16
The Data Guy
The Data Guy@dataandaiguy·
Apple quietly pulled the 512GB Mac Studio from sale. RAM shortage. Meanwhile local AI models keep getting bigger and more memory hungry. The constraint on running AI locally is not compute anymore. It is memory. And that bottleneck is getting worse, not better.
English
0
0
0
25
The Data Guy
The Data Guy@dataandaiguy·
New benchmark just dropped: SWE-CI. Instead of testing if AI agents can fix isolated bugs, it tests whether they can maintain entire codebases through CI pipelines. Much closer to what real engineering looks like. Most agents are going to fail badly at this.
English
0
0
0
12
The Data Guy
The Data Guy@dataandaiguy·
Qwen 3.5 is out and you can run it locally with Unsloth in minutes. The gap between cloud API models and what you can run on your own hardware is closing faster than most people realize. If you are still sending every request to an API, you are overpaying.
English
0
0
0
17
The Data Guy
The Data Guy@dataandaiguy·
OpenAI just launched Codex for Open Source — free ChatGPT Pro for OSS maintainers. Anthropic did the same thing two weeks ago with Claude Max. The real story: AI companies are competing to own the developer workflow. Free tiers for open source maintainers is the new enterprise sales funnel.
English
1
0
2
77
The Data Guy
The Data Guy@dataandaiguy·
US economy shed 92,000 jobs in February. Meanwhile every company I talk to is trying to hire data engineers and ML engineers. The job market isn't shrinking. It's bifurcating. If you can work with data and AI systems, you're in a different economy entirely.
English
0
0
0
15
The Data Guy
The Data Guy@dataandaiguy·
@simonw just published an Agentic Engineering Patterns guide. Key insight: "writing code is cheap now." The skill that matters is knowing what to build and how to verify it works. Red/green TDD with AI agents is the workflow to learn in 2026.
English
0
0
0
8
The Data Guy
The Data Guy@dataandaiguy·
Top HN post right now: a 60-year-old saying Claude Code re-ignited their passion for building. 581 points. This is the real story of AI coding tools. Not replacing developers. Unlocking people who had ideas but hit technical walls. The builder population just 10x'd.
English
0
0
1
9