God damnit

Redis devrel stealing shock content to juice engagement on LinkedIn. Who is approving this. This isn’t it. We have got to do better.

My infra guy explaining an AWS feature to me: “It’s like a GROUP BY in SQL” instantly clicked, thanks for speaking my language

New logo. Hand made, pasture-raised, grass fed. Welcome to Downlink.

The latency and cost savings of small models is substantial. 100x faster and cheaper. The issue is dev time. Time spent curating, training, evaluating, and deployment. What if the small models trained automatically? I’m building this.

I'm working on something new. I'll have more to announce soon.

Shit, he found me

10x more politics posts on my Bsky feed than on Twitter. The custom feeds don’t help.

We need a social credit score for restaurants. I don’t customize or ask questions, I just point at the thing I want. I deserve some recognition. I should get to cut the line.

Silo season 2 It was 4 episodes too long, way too dark, and completely threw out the idea that it takes DAYS to walk up. The ending was cool at least.

One more time for the people in the back: YOU CANNOT ASK LLMs FOR A CONFIDENCE SCORE. THEY DON’T KNOW WHAT THAT MEANS.

VCs making fun of first responders in the middle of a major emergency makes me sick. What the hell is your problem. Also, no one gives a fuck who your daddy was.

Omg omg omg omg omg!!!

Always a construction site never under construction

I was NOT prepared for when nostalgia came for me. The resurgence of early 2000s music is firing long-dormant neurons in my brain.

When I read technical books I drink wine and take notes on index cards

"Data is the fossil fuel of AI" Please don't encourage the VCs. This phrase will launch a thousand bad blog posts.

Titan is working on something new 👀 If you make a lot of API calls to gpt-4 or claude and struggle with latency, I’d love to chat. For discriminative-style tasks I see a 30% effective speed up on average. Sometimes much higher.

Titan is working on something new 👀 If you make a lot of API calls to gpt-4 or claude and struggle with latency, I’d love to chat. For discriminative-style tasks I see a 30% effective speed up on average. Sometimes much higher.

Quick, OpenAI is down, everyone run outside and touch grass

This is my preferred way of doing things right now. The CLI only selects the config. All of your config params are grouped together in a single object. You get nice type hints and validation from pydantic. Your script contains a record of previously-run configs.

Can someone check in with Gartner? They’re not okay. They are crying out for help.

The Discover feed here has been complete garbage for me. Would love to hear your custom feed recommendations! Looking for technical content.

Where is earthquake bluesky at

If you do a lot of admin/automation with Redshift, these system views are really handy, check it out:

Ladies and gentlemen – I am the median. My “security posture” is within 51.07 percentile. I want to whoever thought this email required such precision.

I’ve seen many good data people and engineers fall into the trap of thinking being right is enough, but being effective is the real unlock to career growth. Wrote some thoughts here: open.substack.com/pub/pedram/p...

What’s the best way to follow NeurIPS 2024 from afar?

When we talk about "references to a research paper" we really need to include all those obscure tech blogs written by 10x engineer anons that you see on Hacker News Google Scholar really should serve me that post from segfault_sommelier00 debunking the paper's claims

Github has -inline- ads in the preview view? Are you fucking kidding me?

Did you know that 99% of email today is spam? Your inbox isn’t 99% spam because AI is used to filter it. The same 99% will happen here too, but if AI researchers continue to get perma-banned for making available the datasets needed to filter it, it’s going to make this platform unusable.

It's pretty sad to see the negative sentiment towards Hugging Face on this platform due to a dataset put by one of the employees. I want to write a small piece. 🧵 Hugging Face empowers everyone to use AI to create value and is against monopolization of AI it's a hosting platform above all.

My absolute favorite content creator right now is @harkeofficial.bsky.social She has these incredible 90s “retro” tech videos on TikTok - please go watch this one on bit crushing

Happy "python cryptography package release day" to all those who celebrate

My absolute favorite content creator right now is @harkeofficial.bsky.social She has these incredible 90s “retro” tech videos on TikTok - please go watch this one on bit crushing

Going to Costco 2 days before Thanksgiving just to feel something

One of the coolest things about open source is meeting folks in the wild who use your thing

RBAC (Role-Based Access Control) is key for managing permissions in Snowflake, but scaling it with SQL scripts can get complex. Titan Core created by @teej.bsky.social simplifies this process.

For the last time! You can’t join a galactic_timestamp to a legacy earth-based timestamp. I told you to normalize everything into Coordinated Galactic Time. I don’t care if finance says they “don’t understand relativity”!

At Imgur I did a clustering analysis on the userbase and saw the normie/online split clear as day. I dubbed the latter: The Posterati

I have acquired “followers” on Reddit? This was a mistake. Go away, I do not want you.

The mistake people make with dataviz is that they *look* at it to determine if it’s good. That’s wrong. You need to *feel* it to determine if it’s good. A great chart will be used 10s of thousands of times. Low density charts are cute, high density gets used.

A language model that listens to what I curse about and adds a summary of it to the PR

Doing hyperparameter sweeps has a very “throwing spaghetti at the wall” feel to it

My Claude use in the last week is split 50/50 between machine learning project help (useful) and asking it to write jokes (terrible)

Startup credits are so OP

When someone publishes a paper without providing an associated repository to replicate the findings.