Roland Dreier

But what if it's posting something I want to be true

“as someone I know put it: we’re not just going to forget that computers can write code now. We aren’t.”

aredridel.dinhe.net/2026/02/12/t...

Is there anything on the test about believing Reddit screenshots

Yeah - I'm arguing that if the competition is SaaS like ChatGPT Edu, the open/local alternative is probably, if anything, more $$$

Yeah for sure, although the capability to run a 750B or 1T parameter model on your own HW requires a pretty big investment, it's not just “ollama run zai-org/GLM-5” on your gaming GPU

eg to make a web page, I want a model that not only can do HTML/JavaScript, but also is multi-modal and can handle it when I upload a PNG showing what's wrong

“we'll make a small efficient model that's super focused on one thing” sounds good, but in practice a lot tasks are way broader than they seem at first and I think the power of LLMs is that they can get all kinds of references.

My current experience is the opposite - big general models like Opus 4.6 / GPT-5.3-Codex are better for nearly everything (although fine tuning small models can be useful). And context is super important but what matters is checked into my repo as code or markdown, not stored anywhere else.

authenticity is so important these days

My employer pays for OpenAI Codex, Claude Code, Cursor, and Glean for everyone (and probably a bunch of other AI SaaS products I don't care about) and I don't think anyone would say we've fully outsourced cognitive work

Like I said I think I'm missing some subtext here... but it seems to me like there's a lot of space between licensing ChatGPT Edu for your students and faculty, and becoming a ChatGPT wrapper with a football team

I'm a huge free software guy and I agree that there's a really important place for open models for many reasons. But I'd also argue that for 99% of the things people will do with a model, a polished SaaS product ChatGPT is a better value than “here's GLM-5, good luck!”

I'm not sure I'm getting what you're driving at, but wouldn't the same argument apply to paying for Google Workspace vs just building a data center and doing it yourself? You can choose what to buy but you're gonna have to pay for something.

Screenshot of email from developerresources@511.org: Welcome Roland, You have successfully verified your email address, and are now a part of growing group of wonderful people who can use 511's Open APIs.

Transit people are the best 🥰

https://knowyourmeme.com/memes/im-gonna-stop-you-right-there

“If $DJT were in the S&P 500”

Screenshot of an Ed Zitron post: “$30bn raised from a *telethon* of investors. This is insane. Never seen anything like it in my life. Anthropic is scraping every barrel dry, and guess what? This barely covers the $21 billion in orders it has with Broadcom, let alone the projected $12bn in training costs this year”

Oh my bad, turns out it's terrible news for Anthropic because too many investors participated in the fundraising

Screenshot of Anthropic fundraising announcement: graph illustrating “It has been less than three years since Anthropic earned its first dollar in revenue. Today, our run-rate revenue is $14 billion, with this figure growing over 10x annually in each of those past three years.”

Have you done a welfare check on Zitron recently @davidcrespo.bsky.social?

www.anthropic.com/news/anthrop...

A good thread here

bsky.app/profile/dave...

Roland riding in the passenger seat of a driverless Waymo

Ironically had just gotten out of a Waymo ride

Gold Tesla Cybercab stopped in a left turn lane on El Camino Real

Just saw my first Tesla Cybercab in Palo Alto

(it had a steering wheel smdh)

A post from the “data is the new oil” hype cycle of 2021

Language Machines How generative AI systems capture a core function of language Looking at the emergence of generative AI, Language Machines presents a new theory of meaning i...

(not a light read but worth the effort IMHO: www.upress.umn.edu/978151791932...)

I wish @leifw.bsky.social's critique of “Remainder Humanism” could break through - even if it's true, “a machine will never be able to do *THAT*” is such a cramped and ultimately depressing way to look at what it is to be a person

Two co-founders of Elon Musk's xAI resign, joining exodus xAI co-founders Tony Wu and Jimmy Ba have resigned from the artificial intelligence firm they started with billionaire Elon Musk less than three years ago, they said in social media posts.

I don't know, I might consider leaving!

www.reuters.com/business/two...

From the claudexplorers community on Reddit: AI Psychosis and Claude: one person's experience and reflections Explore this post and more from the claudexplorers community

But Anthropic has a chatbot at claude.ai and OpenAI has a coding agent Codex (openai.com/codex/). Obviously they're two separate companies but it's not clear they're all that different?

www.reddit.com/r/claudexplo...

Or some software engineer who just wants to optimize GPU kernels

Elon Musk Wants to Build an A.I. Satellite Factory on the Moon

I've been in my share of all-hands meetings where execs say wild stuff, but this is next level. Imagine being some rocket engineer who just wants to build better cryogenic turbopumps or something

www.nytimes.com/2026/02/10/t...

also because being a software engineer now feels like being a character in some twee work of po-mo metafiction

Investigators wrangled video from Nancy Guthrie’s Google Nest camera out of ‘backend systems’ It’s not yet clear how the FBI collected the video. Experts said in some cases it’s possible to collect data from the complex infrastructure that has enabled cloud-based cameras.

FWIW Guthrie's camera was a Google Nest

www.nbcnews.com/tech/tech-ne...

Man I wish @weisenthal.bsky.social would bring his talents to bluesky