Glenn Ellingson

iOS 26 is a disaster. Stunningly fugly; unreadable blurs under your text; the widget screen (leftmost swipe) is a disaster - they even broke the one usable widget (recent/search apps); controls keep hiding and morphing so you can't find or hit them... I used to blame Jony Ive... who do I blame now?

Correct. Hosting this stuff itself is not very net profitable; that is not the incentive here. Notice that the “announcement” was in the form of a letter to Jordan’s committee. This is just YT caving to a dangerous government to dodge threats of meddling. I worked on this stuff inside a platform

This is actually my field; this piece of reporting is a lazy summary of an interesting paper, albeit one with very real limits - including that it had LLM actors, not human participants. Algorithmic fixes in real world usage can very significantly affect (improve or degrade) our online experiences

One crazy old man here saying “just give me the transcripts, doesn’t anyone read anymore??”

Trump’s Immigration Enforcement: Free The Criminals, Jail The Innocent

The Trump administration's immigration enforcement has revealed itself to be not just cruel, but fundamentally backwards: They're literally freeing dangerous criminals while manufacturing cases against innocent people. And…

I write paper checks AND drive a stick shift... and I'm a young'un, several months younger than you are!

(I also drive an electric car and build systems for ecommerce and internet platforms... it's yes and, not either or!)

No no this is a perfect use of AI for efficiency: this is the most efficient way to tell your customers to go F themselves, you don't care....

AI search engines cite incorrect sources at an alarming 60% rate, study says CJR study shows AI search services misinform users and ignore publisher exclusion requests.

Do you use LLMs for search/answers? Do you have friends who do? It's time to sit down and have an intervention!

Across all vendors LLMS confidently misidentify information (e.g. fabricating sources) more than half the time -- even when "plain" google does not.

arstechnica.com/ai/2025/03/a...

And while I'm at it, another hot take: the Apple Watch SUCKS. I will spare y'all my 15 bullet points on why. TL;DR: after 4 days of constant use (and twiddling, and googling) I still have almost no idea what it's doing most of the time, and it apparently feels the same way about me.

Only figured out that one when the friend eventually jogged up and said "hey, what are you doing?" -- and when I explained, he checked his pockets...

Once, a friend accidentally brought them with him while running, and I spent 20 minutes searching a specific spot in the bushes on the side of a road where FindMe *swore* they were (and I had biked past there earlier, so it was vaguely plausible).

FindMy knew they were "with me" but I walked the whole house 5 times and it couldn't point me directionally, and the sound it eventually deigned to play was so faint that I could tell when I was in the right room but STILL could not pinpoint them. This is my typical experience.

image of airpods on side table, with candy box lid that had been on top of them

Apple's FindMy is both the best AND the worst feature ever. I misplace EVERYTHING so I love this idea SO MUCH but I just spent 15 minutes using it to find my AirPods, which were within arm's reach when I started searching, hidden under a faraday cage -- oops, I mean a cardboard box lid from See's.

Block and ignore, man. Don’t give the trolls oxygen.

Do you have specific performance/reliability/bias criteria in mind? We did, and we met them or we didn't launch (and we "unlaunched" if standards or performance changed). Not perfectly, of course. But there were a lot of subject matter, data, and ML experts making those decisions.

I will also, with respect, challenge the description of Meta's AI detection as "hamfisted." Content moderation at scale without errors is impossible -- whether done with AI or even (given infinite staff and resources) by humans. (Humans disagree with each other and make mistakes too.)

To see Mark Zuckerberg & Nick Clegg now highlighting only false positives -- and ignoring false negatives -- to cozy up to the current political regime is, frankly, pretty disgusting. Pure politicking, IMO

Speaking as someone who was there, I respectfully disagree that Meta overstepped during covid. The tradeoffs are challenging, of course, and reasonable people will disagree on specific policy decisions. But all work was undertaken with clear and explicit weighing of competing principles.

My favorite analogy is: would you eat a package of chips you saw randomly floating past you in the gutter? News quality is important just like food quality. Get it from somewhere you trust.

We built that same intervention at Meta. Tested well. Not sure if it’s still in use

Yeah, our team tried to do the VR meeting thing; novel, kinda interesting once, but nope & we stopped

As a social media safety professional, 💯. Social media, especially open threaded (aka "twiitterlike") social media is COMPLETELY unsuitable for getting news, discussing politics, and especially any form of deliberation. Just have fun!

Totally different system, with different goals, at bsky. Bsky does minimal required moderation in-house and provides really cool tools for others to build (and easily share to other users) custom labeling and feeds. (Maybe 'cause they have 10 engineers not 10K+ like Meta....)

You might be onto something. I worked on Instagram back in the pandemic days... but have never once posted 😅. Honestly, I felt that a lot of people in that company did not really understand their products -- just their team metrics.

Hang in there, you're all doing great! I remember scaling pressures at PayPal loooong ago, when we tracked how many days until the site would be hard down due to organic growth overloading our non-sharded database architecture that we could only scale by "can anyone sell us a bigger server??" (heh)

In fairness I think most republicans want criminals out too. And various sleazebags, cheats, etc. It’s party leaders, faced with narrow margins and fear of political losses, that circle the wagons. Which, screw their lack of backbone. Plus, DJT is an idiot. Our leaders have bent sometimes too

And, when used as a signal, CN (and other fact-checking programs) can be used to evaluate account/source quality, which is a factor that can be used to affect the distribution of *future* contributions from sources that do/do not have a history of high-quality posting

What do I mean by knock-on value? Well, it gets some group of people actively engaged with evaluating and responding, for one. And effectively coaches them (due to the algos used in the product) to produce "answers" that will have broad appeal across diverse audience members

Community notes is a really interesting & great product. But it's not a T&S solution, or a way to materially impact misinformation distribution; the original authors conceed as much. It's too slow and cannot scale to match the production of misinfo. It *can* add value in some knock-on ways, however.

You're doing social media wrong, Tracy. That's not outrageous (although I am getting FOMO, looks delicious!). Post only airport meals. 🙄