Wanted to separate this part from the main conversation: there are also risks (attack surfaces) that the brand new field of AI OPS is identifying, like LLM-jacking. youtu.be/dibZ1itSvM4?...

I've had a first stab at making a list of GenAI ppl on here but it's proving harder than anticipated! X is still the place to be for that content. Let me know if you're interested in being added or know anyone who would be a good addition😎 bsky.app/profile/did:...

Yes, Bluesky is full of ill-informed AI takes. But that doesn’t have to matter at all. The news you need is out there. In addition to the filters recommended here, I recommend … +

Trump's second term is playing out a lot differently than health care companies and their lobbyists hoped for.

My "Map of Nothing" featured in @sciam.bsky.social has been shortlisted for the 2024 Information is Beautiful Awards. Nothing is the new something. Work with @jenchristiansen.com. www.informationisbeautifulawards.com/showcase/712...

AI code assistants may be close to dead weight if the project is one-off roll-you-own. LLMs give me a huge magnifier when my code adheres to major recognized standards, like FHIR.

To my knowledge, this is the first time a model based on ethical pretraining gets SOTA performance. The smallest model fills in an entirely new niche and, once equipped with external memory local solutions can become effective solutions for a wide range of use cases. More to come.

My Birds Canada Birdathon #AlphabetOfBirds continues! *D* is for ducks, obv. *D* is also for dabblers and divers. Behaviour is a big clue in bird ID. Watch for "butts-up" dabblers like gadwall & mallard and "blooping" divers like bufflehead & goldeneye #birds 🌿 www.canadahelps.org/me/6n3u2yyj

I now believe that every programmer should pair program with an LLM. That's the way to think of it -- you're pair programming with a 4.0-getting recent college graduate who has no life experience. It's a pleasure to ask, "What's the best way to debug this issue?" and get a page of code in 5 seconds.

Tariff anecdata: in 2024, I got an inquiry once a week from a home office looking to invest and wondering what I was working on. Now: 3 to 5 inquiries per day. The spike started a few days after the tariff announcement. If the stock market's a gamble, they'd rather gamble on something else.

A tip I can offer to improve the quality of code produced by LLMs: they guess the next line; they don't ratiocinate. Breadcrumbs are better than explanations. Genericize any functions that might benefit. Build with ridiculous type safety. Give a DO and DON'T example in the comments on each function.

Anthropic foresees AI virtual employees soon, with their own accounts and roles. Securing these AI identities and managing their network access is a crucial, yet unsolved, cybersecurity challenge. There is a strong need to keep an eye out for misuse. 🩺💻 #MLSky

This is a known business problem for security-as-a-service companies like Snyk. There aren't many real vulnerabilities anymore. Snyk has to find new ones to justify subscription price, causing library maintainers headaches, but the vulns they find are minor.

"Crypto" as slang refers to a lot, but formally, there's a separate notion: "community currency." It's controlled, so it can't be rug-pulled. I've discussed creation of a couple currencies in the medical space. Hadn't thought of this. Happy to discuss architecture if someone wants to motor it.

Happy Easter everyone! #wildlife

Many Asian creators right now are re-making machine-generated content with people only. Musically most impressive to me is Singapore-based husband and wife team Dharni and Mamiko's a capella covers of turn-of-the-century Eurodance. Here's their take on Tiesto's remix of Barber's Adagio for Strings

Final thought: getting high-quality results out of genAI is a complex management problem. When people talk about "AI slop," what they are really saying (even if they don't realize it) is the person using the tool is a sloppy manager. Understandable, as so few receive training in those skills.

I decided to say please and thank you to Claude because I didn't like the person I sounded like when I spoke declaratively, and I didn't want to lose social cadence. UK maths explainer Hannah Fry has a different take on why to be polite to AI. youtube.com/shorts/rVlmb...

I believe we will see question-asking emerge as key professional skill. Training in that skill has been limited to leadership. Requires technical acumen and self-discipline. IME low-performing leaders are less willing to ask for time/task breakdowns: they are afraid to be held to the same standard.

The Trump administration revoked Biden's 36-page responsible AI executive order, replacing it with 7 paragraphs. Now, new memos fill in the blanks on how the U.S. govt is supposed to use AI. There are lots of differences in the new approach—@statnews.com's @caseyross.bsky.social & I break it down:

Some people may be unable to use a mouse, and instead rely exclusively on keyboards, or use assistive technologies such as speech recognition, head pointers, mouth sticks, or eye-gaze tracking systems. If a website is accessible only to mouse users, these people will be excluded.

Scottsdale police say crypto ATMs are becoming hotbed for fraud www.azfamily.com/2025/04/14/s...

Let’s not underestimate what a significant step forward around the Pandemic Agreement we saw early this morning. Multilateralism won out. There’s a way to go but gees given the global health turbulence and potential biomedical threats emerging this is an important win. We needed this.

Just sent a feature request to Anthropic to offer HIPAA Business Associate Agreements on their Max plan.

"Last year, the state chancellor’s office estimated 25 percent of community college applicants were bots." It seems fake students are trying to steal financial aid funds. They are often using AI programs to generate classwork

Reported hack of 4chan, including names, email addresses and IPs of users. Some mods used .gov emails apparently.

As someone who used to maintain a somewhat successful open source project and has been burned pretty badly by people using, I can more than relate to this take explaining what open source isn't. Open source does not mean... 👇 vale.rocks/posts/open-s...

Nobody starting a career in STEM in 2025 should skip learning about AI. Take a formal class if you can, but anyone can reinvent themselves to work with AI intelligently. I have some suggestions for how to teach yourself, even if your math is limited: www.thisgenomiclife.org/p/how-to-tea...

MORE 👏 REPORTS 👏 DO 👏 NOT 👏 ALWAYS 👏 MEAN 👏 MORE 👏 ABUSE When stats like this go up, it usually means that the platform has gotten much better at detection and reporting grooming. News like this does a disservice to everyone just bc platforms invest in tech / teams www.rnz.co.nz/news/nationa...

New blog post: I'm publishing a realistic AI timeline. A fully speculative exercise going all the way till AGI but still grounded into current research and the coming end of pretraining as we know it. And with a radical different set of premises. vintagedata.org/blog/posts/r...

New Research paper our about AI inclusivity for neurodivergent folks link.springer.com/article/10.1...

Fifteen minutes ago, I got a phone call that the residence program I built the financial model for, just moved in their first residents. It feels good to have a win after so many people have faced cuts or layoffs. Do not doomscroll. Keep working in the real world. People want to help.

If you have expertise in the design and evaluation of digital health apps, please consider joining the IEEE Global App registry project, which will evaluate therapeutic health apps in terms of clinical efficacy, usability, & technical safeguards. standards.ieee.org/products-pro...

My whole FYP on TikTok is factories in China advertising their products after deciding they want to sell direct to consumer.

I've said it before, and I'll say it again: the implementation of policy depends on technology. If the technology fails, the policy fails. Sure, you can declare that tariffs are 140% (or whatever they are today, I cannot keep track), but if the CBP software doesn't support that, then they are not!

This strongly correlates to my own experience. Using an AI Code Assistant feels more like technical management than programming. I see little need for coders/contractors/offshore shops who fulfill exactly what they're told to do. Lots of need for people who can write requirements and evaluate.

This is a remarkable series of cautionary tales. We use AI code assistants a lot, and I'm going to assign this article as required reading for new hires: example after example of what not to do. h/t @johnfallsopp.com aftermath.site/ai-video-gam...

Bay Area folks: Join us at Palo Alto VC Nest on April 10, 430–630PM for a panel on Agentic AI in Healthcare—real talk on the challenges of implementing AI in health systems. Clinicians, engineers & investors welcome. Guaranteed fun b/c @mattsakumoto.bsky.social & I will be there 😜 lu.ma/dfhb85a1

I’m finding it really difficult to publish pieces critical of large language models or AI in medical journals. Some of our papers show hard evidence of why the models are not ready for prime time. But journals seem to prefer the “AI is a doctor!” papers. www.nature.com/articles/s41...

It was a pleasure illustrating your paper! Congratulations on the amazing research!

Where Are They Now: Here's what Antoine Dodson, also known as the 'Hide Yo' Kids, Hide Yo' Wife' Guy, has been up to since his viral 'Bed Intruder' video and song. Read more here 👇

This Starter Pack is a list of about 50 nature photographers. Most of them post primarily nature although there will be some other topics as well. Following all of them will give you a constant stream of animals and other nature on your timeline.

Over the past few months, I’ve been tracking and curating a database of AI-involved drug candidates, 380+ so far across 88 companies. It’s a mix of repurposed drugs, AI-designed molecules, and smarter clinical trials. (Link to database in article) www.aiinlabcoat.com/p/what-makes...

AI chatbot somewhat more effective than waitlist control. An effect of d=0,61 is something you would expect from a self-guided intervention, but I suspect it’s much more expensive.

Trump administration sued over decision to rescind billions in health funding www.statnews.com/2025/04/01/t... via @statnews.com

OpenAI plans to release a new 'open' AI language model in the coming months | TechCrunch https://techcrunch.com/2025/03/31/openai-plans-to-release-a-new-open-language-model-in-the-coming-months/ #AI #OpenSource

This article has now been updated after a conversation with ThinkingOne, who released the latest X file. #infosec #X www.forbes.com/sites/daveyw...