The Age of Twitter is finally ending. Can Substack take its place?

Some thoughts on the future of the media.

Twitter is dying. What social media platform will take its place?

Blinking cursors are annoying and useless. Unless some information has changed that I need to be aware of, nothing in the UI should move. In the rare event I lose track of where the cursor is I'll just move it back and forth with the arrow buttons. Blinking the cursor serves no other purpose than distracting me and it's unfortunate that it ever became a UI convention.

I just read about gender studies and sexual orientations of ancient Vikings and I actually enjoyed it. Does this mean I’m woke now?

It’s a strange feeling scrolling back through my follow history to see how many of them that used to be super active in the very early days have gone silent. This community has changed a lot. We’ve definitely become less dev centric and things “just work” most of the time unlike the early days. Eventually most of the pubkeys will stop sending notes, we will come full circle, and it’ll just be @QW, @Derek Ross, and I hanging out in a global feed telling newcomers how to set their lightning address, post an invoice and tag other accounts, alongside @jack throwing out shakas to random notes by newcomers.

Daring Fireball

Trump Threatens Apple With 25 Percent Tariffs on iPhones Assembled in India

Link to: https://truthsocial.com/@realDonaldTrump/posts/114556874484491575

Trump’s obsession with making iPhones in America gets more nuts when you look at the numbers. Apple sells 150k iPhones on an average day. In a launch week they’ll sell millions a day. The insanely gigantic city sized factories and living facilities needed to pull that off would simply not work in the US. We don’t have the large amounts of low cost labor it would require.

Trump Administration Says It Is Halting Harvard’s Ability to Enroll International Students

The move was a major escalation in the administration’s efforts to pressure the college to fall in line with President Trump’s demands.

This is absolutely nuts. The Trump admin forcing Harvard to stop letting in foreign students and making all foreign students studying at Harvard move to a different university. In a PhD program with your advisor and research almost done? Sucks to be you, good luck replicating everything somewhere else. Even the most natalist frothing at the mouth America firster who never wants to hear a word of Spanish spoken in their country has to admit that wealthy and highly intelligent young immigrants coming to the US and paying top dollar at American universities and subsidizing domestic students is a great thing for America.

The whole Bitcoin Core op_return fiasco has laid bare how horrifically centralized the Bitcoin ecosystem really is. Bitcoin Core is the only really viable implementation of Bitcoin. This may be the biggest threat to Bitcoin right now. We need different node implementations in different languages with different architectures made by totally different people. The fact that the only real alternative to Bitcoin Core is Bitcoin Knots which is just a fork of Core except with one guy yoloing commits should be embarrassing. This is the biggest threat to Bitcoin and the only solution I see is for companies which rely on Bitcoin to sponsor developers to work on different implementations. That’s my rant for the day.

Google DeepMind

Gemini Diffusion

Gemini Diffusion is our state-of-the-art research model exploring what diffusion means for language – and text generation.

This is really exciting stuff. Google released a diffusion language model. The speeds you can get with this type of architecture are absolutely nuts. It's also good to see AI labs experimenting with different architectures. I have worried for a while that perhaps AI development has fallen into a local minimum and we didn't investigate other approaches deeply enough. Perhaps Transformers aren't the absolute best architecture for language models. But if you iterate on a substandard solution long enough it might be possible to outdo other approaches that won't be as good initially but would eventually lead to much better results if pursued long enough. Narrowing the search space too soon and only focusing on Transformer-like architectures might mean we might miss out on better solutions. Trying other architectures expands the search space.

Anthropic just dropped Claude 4 which includes the new Opus! I thought we would never see another Opus after they didn’t release a version 3 or 3.5.

Introducing Claude 4

Discover Claude 4's breakthrough AI capabilities. Experience more reliable, interpretable assistance for complex tasks across work and learning.

After the Trump admin and several conservative media personalities who claim to have viewed the file on Epstein have all said that he did in fact just kill himself, it seems odd to me that so many right wingers decided to do a real life version of “the conspiracy goes higher than I thought” meme and decide that it’s more likely that all these other people are also in on the coverup rather than update their beliefs to assign a much lower probability that Epstein didn’t kill himself. Maybe everyone is in on the coverup, but you gotta admit the simpler explanation is just that Epstein really did kill himself. https://www.reddit.com/r/Jokes/comments/b7bhdu/a_jfk_conspiracy_theorist_dies_and_goes_to_heaven/

The Babylon Bee

Oops: Pastor Accidentally Includes ChatGPT Prompts From When He Asked It To Write His Sermon

IRVINE, CA — Local Associate Pastor Mark Sullivan took a leave of absence following a disastrous Sunday service in which he presented a message t...

For some reason it has only just now occurred to me that pastors are probably using ChatGPT to write their sermons.

When the introvert designs the elevator.

Google

Gemini 2.5: Our most intelligent models are getting even better

At I/O 2025, we shared updates to our Gemini 2.5 model series and Deep Think, an experimental enhanced reasoning mode for 2.5 Pro.

micahflee

DDoSecrets publishes 410 GB of heap dumps, hacked from TeleMessage's archive server

💡Update May 21, 2025: I've written a new detailed analysis of the heap dump dataset itself: TeleMessage customers include DC Police, Andreessen ...

Signalgate just keeps getting worse. Apparently the company that sells the Signal fork the Trump administration was using left up a public endpoint where you could capture heapdumps of their servers and see messages in plain text.

The Israeli singer at Eurovision got cheated out of her win. So many of the country’s juries gave her zero points just to protest Israel. Her performance and song was so much better than the first place finisher. Also, what a story. She had to hide under a pile of dead bodies at the Nova music festival until she could be saved.

CNBC

Coinbase says hackers bribed staff to steal customer data and are demanding $20 million ransom

The incident may cost Coinbase up to $400 million to fix, the company estimated.

Sooo… now feels like a good time to remind everyone that if you don’t own the keys, you own the coins. You don’t really own the coins on coinbase, you own an IOU from coinbase.

Marginal REVOLUTION

The new FTC commissioner Mark Meador - Marginal REVOLUTION

Frankly, he is just flat out terrible.  You can read his recent document here.  Early on he tells us: Conservatives must reaffirm that concentrat...

Wow great Trump just appointed the MAGA version of Elizabeth Warren to be the new FTC chair.

arXiv.org

LLMs Get Lost In Multi-Turn Conversation

Large Language Models (LLMs) are conversational interfaces. As such, LLMs have the potential to assist their users not only when they can fully spe...

Microsoft wrote a paper about how much LLMs degrade in multi-turn convos. I’m glad my hunch was finally confirmed. I’ve felt for a while that resetting the context and tweaking the first prompt is often far more effective than clarifying with a follow up prompt when the LLM misunderstands or fails to solve the problem on the first attempt. My guess was that there just aren’t enough examples of multi turn convos in the RLHF training dataset, especially of LLMs recovering from a mistake, but I hadn’t considered how benchmarks play into it. Because the format of most benchmarks is just “question->answer” without any follow ups your LLM can test well and still fall apart the moment the user asks a second question.

To the surprise of literally no one, once you start randomly imposing tariffs on everyone you get really unpopular.

US popularity collapses worldwide in wake of Trump’s return – POLITICO

Google’s latest version of Gemini 2.5 Pro is freakily smart. I was going to send an email pointing out where I thought someone was incorrect about the ability of moral facts to affect our brains backed up with research. I fed the email through Gemini 2.5 Pro to review it and after several back and forths became convinced that I was actually incorrect and completely misunderstanding the core of the argument.