iiinova — where your digital self keeps evolving

Reposted by Mark Torres

William J. Brady @williambrady.bsky.social · 27/05/2026

As I mentioned in the below thread, this project involved many feats of engineering, led by the fantastic @markptorres.bsky.social. If you're a CS or CSS person interested in the gory details, see his blog post: markptorres.com/research/202...

markptorres.com

How we built the infrastructure for a large-scale social media field experiment during the 2024 US election

What we built

0113

Reposted by Mark Torres

William J. Brady @williambrady.bsky.social · 27/05/2026

✨New paper out @nature.com ✨ For 8 weeks around the 2024 US election, we randomly assigned 2,000 people to use social media algos we built ourselves. Do engagement-based algorithms amplify intergroup, moral & emotional (IME) content—and does that distort how we see political norms? 🧵🔗 👇

513762

Reposted by Mark Torres

William J. Brady @williambrady.bsky.social · 27/05/2026

Special shoutout to @markptorres.bsky.social my senior lab engineer. It felt like we ran a start-up for a year...If you're a CS or CSS person interested in the gory details of engineering that went into this ambitious project, we are doing a separate thread for you! See my profile

171

Mark Torres @markptorres.bsky.social · 25/05/2025

Claude 4 is the first LLM that has allowed me to actually "vibe code" a decently complicated app in Cursor purely through instructions and markdown files and without having to write a single line of code. Had to intervene a few times in the chat but otherwise really impressive!

010

Mark Torres @markptorres.bsky.social · 16/03/2025

I can't believe that in 2025, we can run reasoning models locally. I finally got to try Ollama and QwQ and it's really impressive. Next step is to set up Ollama + Cursor. Can't imagine where things will be in 2026 and beyond. ollama.com/library/qwq mem.ai/p/bf6ew6HSm1...

ollama.com

qwq

QwQ is the reasoning model of the Qwen series.

030

Mark Torres @markptorres.bsky.social · 31/01/2025

I still think people should step back sometimes and just think about how far AI has come in the past 5 years. NLP used to be "fine-tune BERT and hope it works" to "do one-shot inference, on any task, using GPT 4o-mini". Can't take it for granted that SOTA AI is an API call away...

030

Mark Torres @markptorres.bsky.social · 31/01/2025

The reasoning trace of OpenAI's o3-mini seems like them trying to strike a balance between "we want to keep our reasoning traces IP" and "we want people to think we're being transparent". Still definitely prefer the depth of DeepSeek's traces, though it's still too early to tell.

010

Mark Torres @markptorres.bsky.social · 17/01/2025

I just read Stolen Focus and I really recommend it to anyone interested in a holistic systems overview of why it’s so hard to keep your attention on anything. Who could’ve guessed that the key for success is eating healthy, drinking water, sleeping 7-8 hours, exercising, and reading books 🤣

goodreads.com

Stolen Focus: Why You Can't Pay Attention— and How to T…

Our ability to pay attention is collapsing. From the Ne…

030

Mark Torres @markptorres.bsky.social · 17/01/2025

Self-identifying as some variant of “I’m a critical/free thinker” or “I can think for myself” is almost certainly a signal that one cannot, in fact, think critically for themselves.

000

Mark Torres @markptorres.bsky.social · 17/01/2025

Heard this zinger take at a talk: “Most lay people shouldn’t read scientific papers, even if they think they can, because most people don’t understand that science is an iterative process. There’s no “right answer”, and people do disagree. Even laws are just ideas that we haven’t proven wrong yet.”

100

Reposted by Mark Torres

Casey Newton @caseynewton.bsky.social · 15/01/2025

NEW: Meta has quietly dismantled the system that prevented misinformation from spreading in the United States. Machine-learning classifiers that once identified viral hoaxes and limited their reach have now been switched off, Platformer has learned www.platformer.news/meta-ends-mi...

Behind the scenes, the company was also quietly dismantling a system to prevent the spread of misinformation. When the company announced on Jan. 7 that it would end its fact-checking partnerships, the company also instructed teams responsible for ranking content in the company’s apps to stop penalizing misinformation, according to sources and an internal document obtained by Platformer.

The result is that the sort of viral hoaxes that ran roughshod over the platform during the 2016 US presidential election — “Pope Francis endorses Trump,” Pizzagate, and all the rest — are now just as eligible for free amplification on Facebook, Instagram, and Threads as true stories.

1360260129292

Mark Torres @markptorres.bsky.social · 05/01/2025

Overall really good! Great way to digest research papers, plus I've been running out of new podcast episodes lately so it's nice to be able to make my own custom podcast episodes. Wish I could steer the podcasting behavior a little more and wish that it were longer but I'm liking it so far.

000

Mark Torres @markptorres.bsky.social · 05/01/2025

More NotebookLM notes: - It has a funny pronunciation of "SQL" that I've never heard before (almost like "sekl"?) - The two podcast hosts are always the same and I can only mildly steer their behavior with system prompts. - There's weird times where the hosts like to finish each other's sentences?

100

Mark Torres @markptorres.bsky.social · 05/01/2025

I've been experimenting with NotebookLM to read papers in podcast form and it's been great at it! If I add more than 1-2 papers though, I find that the quality suffers. Plus it caps out at ~20 minutes, can ramble, and its adherence to system prompts is iffy. Great tool though!

notebooklm.google

Google NotebookLM | Note Taking & Research Assistant Powered by AI

Use the power of AI for quick summarization and note taking, NotebookLM is your powerful virtual research assistant rooted in information you can trust.

110

Mark Torres @markptorres.bsky.social · 11/12/2024

The Great Fire of Rome happened when the data centers full of the latest 3,000 nm chips caught fire and there weren’t enough aqueducts to cool them down. Completely unrelated to Nero joining AMD’s board just 6 months before and sitting on Palatine Hill with Lisa Su to watch NVIDIA burn.

010

Mark Torres @markptorres.bsky.social · 11/12/2024

Can you imagine the amount of aqueducts they must've had to build to cool down all their data centers? Back then, Nvidia must've been on their 5,000 nm chips, so hopefully the Romans and Greeks called in ahead to reserve the 4,000 nm chips in advance.

110

Mark Torres @markptorres.bsky.social · 10/12/2024

I think that's true. I suppose the caveat is that Mookie plays one position for weeks or months at a time, whereas it did seem like you'd know where Zobrist was playing only when the lineup card came out. The Rays seem to like generic IF and OF players instead of by position, especially post-Longo.

010

Mark Torres @markptorres.bsky.social · 10/12/2024

I wonder if there's anyone who conclusively out-Zobristed Zobrist himself over the course of multiple seasons. Zorilla was a cog in some good Rays and Cubs teams before being a super-utility player was cool.

110

Mark Torres @markptorres.bsky.social · 10/12/2024

I'm not too aware of AI detection research but this was an interesting way to do it. It's trivial to fool normal AI checkers, since plain zero shot fails. But you can build a better AI checker if you include a retrieval step comparing a text to known AI-generated text.

dl.acm.org

Paraphrasing evades detectors of AI-generated text, but retrieval is an effective defense | Proceedings of the 37th International Conference on Neural Information Processing Systems

000

Mark Torres @markptorres.bsky.social · 10/12/2024

I wonder if filtering spam in the age of LLMs is similar to designing good CAPTCHAs now, where it's hard to create a filter that catches the best LLMs but is also easy enough for the average person. Especially true since it's hard to reliably tell LLM-generated text from human text.

100

Reposted by Mark Torres

Mark Torres @markptorres.bsky.social · 10/12/2024

another test post

001

Reposted by Mark Torres

Mark Torres @markptorres.bsky.social · 10/12/2024

test post 4

001

Reposted by Mark Torres

Nick Fisher @hydroxide.dev · 09/12/2024

Oh wow, LG just released their own open source* LLM. If their published benchmarks are accurate, the 32B model is at least on par with Qwen2.5 (which is already an incredibly strong model), if not better. www.lgresearch.ai/blog/view?se... huggingface.co/LGAI-EXAONE * open weights

lgresearch.ai

Open-sourcing Three EXAONE 3.5 Models : Frontier-level Model, Top-tier Performance in Instruction Following and Long Context Capabilities - LG AI Research BLOG

3152

Mark Torres @markptorres.bsky.social · 09/12/2024

I keep getting ads for "OpenAI pays its LLM engineers 750k, here are 7 projects to get YOU an LLM engineering job", what absolute slop. Engineers also sometimes forget that they're hired to solve problems that happen to use code, so we can't forget what those problems are in the first place.

020

Mark Torres @markptorres.bsky.social · 09/12/2024

Oh man, I thought this was just me, my entire app runs entirely on several of these bash scripts LOL

github.com

110

Mark Torres @markptorres.bsky.social · 09/12/2024

Probably what went through his head after it went down:

130

Reposted by Mark Torres

🎃Darin S Pumpkins🎃 @darinself.com · 09/12/2024

I'm not mad at a baseball player getting paid his money, but its wild to me that MLB has teams that can shell out over $700 million for a player and teams that apparently can't build a stadium without taxpayer money.

515434

Mark Torres @markptorres.bsky.social · 09/12/2024

If he were on-call this past week and more pipelines broke in prod, none of this would've happened smh

010

Mark Torres @markptorres.bsky.social · 09/12/2024

I finally learned what Snowflake and Databricks actually do and I now question why I worked for 3 years building essentially an in-house, worse version of what someone with basic SQL knowledge could have done on Snowflake...

000

Mark Torres @markptorres.bsky.social · 09/12/2024

The news just came out about the arrest of the CEO's killer and Polymarket is wayyyyy too quick with releasing their latest betting odds 😂

000

Mark Torres @markptorres.bsky.social · 09/12/2024

I think they match well with someone like Seattle, they’ve got too much pitching and need some hitting. Mets should gun for someone like Woo or Hancock, that goes a long way towards making that team more complete. Also would be good for better luck with health, Marte needs to stay on the field

010

Mark Torres @markptorres.bsky.social · 09/12/2024

Yeah I agree, and I think if they get one more big star and then get a few more pieces like they did last year with Manaea, Severino, and Iglesias, they’re going to be a top contender. Problem is if they don’t, that lineup starts looking awfully top heavy like the Yanks or KC.

100

Mark Torres @markptorres.bsky.social · 09/12/2024

No point for Lindor and Soto to get on base if nobody is going to knock them in 🤣 as top-heavy as the Yankees lineup was, this Mets lineup isn’t that much better. Maybe if they get Alonso back, then things change.

110

Mark Torres @markptorres.bsky.social · 09/12/2024

I use custom meta-prompts for ChatGPT to have it write the way I want it to write. I pasted in the last 15 paragraphs worth of ChatGPT responses from my latest ChatGPT thread into GPTZero and it marked it as 72% human…

010

Mark Torres @markptorres.bsky.social · 09/12/2024

The stupid thing about AI detectors is that none of them actually work. It’s trivial to break one. You can just as easily ask ChatGPT “write in the style of [insert author] and write sentences in a clear, declarative manner” and pass any AI detectors. But people love peddling the snake oil

180

Mark Torres @markptorres.bsky.social · 09/12/2024

At best probably 5% of people I know outside of tech know what ChatGPT, and probably 95% of that subset hasn’t tried it because “doesn’t it just make things up?” and “I went to school and I’m smart, I don’t need AI’s help”.

100

Mark Torres @markptorres.bsky.social · 09/12/2024

OpenAI was happy to operate as a nonprofit until they realized they actually were sitting on a goldmine worth billions of dollars, and then suddenly we see “oh but we have to fulfill our fiduciary responsibilities to our investors”

100

Mark Torres @markptorres.bsky.social · 09/12/2024

From the looks of it, OpenAI is just getting started: “[OpenAI’s] blueprint also outlines a North American AI alliance to compete with China's initiatives and a National Transmission Highway Act "as ambitious as the 1956 National Interstate and Defense Highways Act."

cnbc.com

OpenAI to present plans for U.S. AI strategy and an alliance to compete with China

OpenAI's official blueprint for U.S. AI infrastructure involves AI economic zones and government projects funded by private investors, according to a document.

000

Mark Torres @markptorres.bsky.social · 09/12/2024

All of this to play in a division with the Phillies and Braves and then to get bounced in the playoffs by the Dodgers 🤣 Mets hardly have a rotation behind Senga and Petersen and LA’s slotting Kershaw and Gonsolin as their #5 and #6. Money would’ve been better spent in the AL.

010

Mark Torres @markptorres.bsky.social · 09/12/2024

I’ve found it helpful to ask ChatGPT things like “why couldn’t I do it like this [insert steps]?” or “explain how they came up with that way of solving the problem. What wasn’t working before and why would an idea like this work?”. It was pretty helpful for understanding why transformers work.

110

Mark Torres @markptorres.bsky.social · 04/12/2024

Trying to learn more about the Bluesky API! Easiest way is just replying to yourself 🤣

000

Reposted by Mark Torres

Mark Torres @markptorres.bsky.social · 03/08/2024

reply to my own post!

111

Mark Torres @markptorres.bsky.social · 02/12/2024

I've never liked tools that try to be "AI writing assistants", but I do like asking ChatGPT to analyze what I've written, give me detailed critique, and then give me line-by-line suggestions for how to improve clarity. Hard to make a tool though that works for everyone's style and use case.

010

Mark Torres @markptorres.bsky.social · 01/12/2024

Yeah this is a problem I’ve had because most political posts are about current events. Works well normally, but errors on, say, if there’s a new politician/person/law/bill in the news and they’re not in the knowledge base, which isn’t rare. I’m tweaking with some RAG-esque ways to get context for it

000

Mark Torres @markptorres.bsky.social · 30/11/2024

I didn’t know LLMs knew stuff like country-specific politics! Wonder how it does if you asked it to label political parties. I use LLMs to classify Democrat/Republican posts for US politics and it works well, and for language I use fasttext, which can label a batch of millions of posts in seconds.

huggingface.co

facebook/fasttext-language-identification · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

110