i love giving entities that are more gullible than my grandma access to both my full home infrastructure and unfiltered external communication methods

From now on we are required by law to call them fair use machines

Not sure there is a clearer tell that AI *users* and their *personal* and *work* data are the product, not the Artificial Intelligence-Ignorance LLMs themselves. www.bloomberg.com/news/article...

I've been trying to articulate why the fawning, complimentary responses from AI chatbots feel so insidious to me. I've finally figured out how to explain it. Wrote a long piece on how current model training and design choices threaten our critical thinking skills: maggieappleton.com/ai-enlighten...

Don't leave AI to the STEM folks. They are often far worse at getting AI to do stuff than those with a liberal arts or social science bent. LLMs are built from the vast corpus human expression, and knowing the history & obscure corners of human works lets you do far more with AI & get its limits.

🚨 New paper from us: Given they are trained on human data, can you use psychological techniques that work on humans to persuade AI? Yes! Applying Cialdini's principles for human influence more than doubles the chance of GPT-4o-mini agreeing to objectionable requests compared to controls.

Elon saying he’s going to reprogram his chatbot’s view of history is a perfect example of why these kinds of AI products aren’t the neutral “tools” people always defend them as. They’re little political projects reflecting the ideologies of their creators

Got distracted today & did a little experiment on ChatGPT 4o. I told it I had a question about AI & asked it to suggest six experts to contact. In one condition, I put "tough-minded and rigorous" before "experts". In the other condition, I put "friendly and kind" before "experts". ⬇️

Not good. Thank you to @wired.com for documenting and to the researchers for speaking up.

I feel like my ability to figure out whether someone knows their stuff when they're talking about AI has become so sharply honed the more time I spend switching contexts between technical and non-technical people

Anthropic’s MCP should be more widely used in enterprise

An X engineer posted this output from Grok to demonstrate how "good" their LLM is (CW: racism)

The left version of ‘run government like a business’ and ‘cut red tape’ is that there are a ton of asinine constraints that the public sector is under, things that would instantly be recognized as insane practices in a well run private firm, in the name of ‘small government’.

This is sort of the skeleton key to understanding a lot of shit: there 100% is a non-trivial amount of inefficiency and waste in the federal bureaucracy, and most of it is a *direct result* of ‘reforms’ that are meant to assuage the people who complain about “waste and inefficiency in govt”

The most effective responsible generative AI technique is to proactively gatekeep LLMs from decision-makers who don't know how transformers work

A lot of people don't know how useful Model Context Protocol is - easily Anthropic's most underrated feature. Allows Claude to use your computer and applications with natural language. Official and unofficial MCP servers, which are like extensions, here: www.mcpservers.ai

If you have to prompt engineer confidence scores, then you already know they're not going to be reliable. arxiv.org/abs/2412.14737

If you, like me, have a feeling of despair reading this news, I'd ask that you consider donating to the organizations who do this work: Global Fund to Fight AIDS, Tuberculosis, and Malaria: act.unfoundation.org/FJvB3vUCJUep... International Medical Corps: internationalmedicalcorps.org?form=Main

It's actually theft of taxpayers money when websites go dark. We paid for that data

okay so what legal authority does elon musk have to commandeer the office of personnel management and access reams of sensitive data? www.reuters.com/world/us/mus...

Today, we are publishing the first-ever International AI Safety Report, backed by 30 countries and the OECD, UN, and EU. It summarises the state of the science on AI capabilities and risks, and how to mitigate those risks. 🧵 Full Report: assets.publishing.service.gov.uk/media/679a0c... 1/21

positron is the best IDE for data science and it's been my daily driver for over a year

Another takeaway is that we've been Stockholm Syndrome'd to think pandas has a reasonable and intuitive API for data analysis

middle managers who've never written a single line of code or built an ml model before

Every time I visit the North End here in Boston, I'm grateful to not be swallowed up by a flood of liquid sugarcane extract

A strange disconnect I've been thinking about: in the 1940s, 2.5m Indian soldiers fought against fascism in WW2. Many of their children came to America in the 90s tech wave. Now some of their kids and grandkids support the same ideologies they fought against

#Boston friends: want to see how the city's services are measuring up? I built CityUpdate Boston - daily plain-language summaries of municipal performance data, from EMS response times to sign repairs. Check it out: hersh-gupta.github.io/cityupdate

I gave deepseek-r1 (q8_0) a math problem and it got there after 10 minutes of non-stop trial and error

The forceful intrusion of Copilot, OneDrive, etc from Microsoft at the OS-level all but guarantees that I'll be using Linux from here on out

Researchers: this is _not_ how you evaluate LLMs www.nature.com/articles/s41...

Public sector orgs with large compute resources keep getting suckered into cloud data contracts because nobody on staff knows what kubernetes is - what a waste of taxpayer money

To deal with cobras, the British Raj placed a bounty on them, which worked at first, but resulted in more cobras as people bred them to collect the bounty The US bans TikTok for China ties, so users flee to Xaiohongshu "little red book" which sounds a lot like Mao's Little Red Book

Read the blog post: huggingface.co/blog/ethics-... Brillant work by @mmitchell.bsky.social @evijit.io @sashamtl.bsky.social @giadapistilli.com

@pahlkadot.bsky.social's observations about hiring in government match my own experience and this Odd Lots episode is a great listen, but I'm not sure who at Bloomberg was responsible for the overly editorialized title found on their website

Thomas Malthus:

The opposite of code smell: code fragrance youtu.be/I72uD8ED73U?...

Rebranding all models/algorithms as "AI" for headlines isn't technically wrong but it's poor journalistic practice that deserves more scrutiny than is currently given

Maybe it's too early to tell but AMD missed the opportunity to bifurcate AI prosumers from gamers with something similar to Nvidia's Digits, but the Ryzen AI Max Pro+ seems undercooked in comparison to the GB10

People really struggle with the concept that you can be excited about AI technology while also thinking critically about its implications.

there's a LOT of magic in FastAPI 😅

NotebookLM's podcast format is good, but could you imagine if the AI hosts had the personalities and rapport of Kara Swisher and Scott Galloway?

the universe if pytorch supported amd on windows

Has anyone tried to create an LLM agent that generates @evidence.dev apps/pages, given data sources and instructions/queries?

Robust academic research is crucial for human-centered AI. At our next seminar, Stanford HAI Executive Director Russell Wald will discuss why governments must boost investment in public sector AI research and propose policies to balance the AI ecosystem: hai.stanford.edu/events/russe...

A lot of folks might not realize that Tableau and Power BI are soon-to-be legacy software, and data analysts would be better positioned to learn modern tools like evidence.dev #DataViz #databs

Somehow got ollama working with my laptop's unsupported AMD GPU using unofficial DLLs on Windows. Big ups to this repo author: github.com/likelovewant... Also, why can't AMD just make it easier to use ROCm?

Seeing uvx, markitdown, and llm being used in a *single* line to invoke the tool, have it parse online video content, and then sending that to a language model to be summarized is really something else

City 311 employees spend a lot of time parsing service requests. To make this easier, I built a proof-of-concept that uses DSPy and vision models (like LLaVA) to augment 311 workflows: 📝: hershgupta.com/posts/smarte... 💻: github.com/hersh-gupta/...

I only just found out that DSPy has an image adapter implementation for vision models?? This kind of functionality is exactly what I needed, but not a mention of it on the dspy.ai website?