This was the week when I tried every single commodity LLM, and a few other things we’re labelling “AI”. And let me tell you, oh boy have my opinions on AI flip-flopped like the integrity of a career politician. More on that after the break.

This week in data:

• Papers read: 3
• Distance run: 13.29km
• Gigs attended: 1

To put it mildly, it really hasn’t been a good week from a current affairs perspective. On Thursday I started writing a message to someone and by the time I sent it on Sunday I had to redraft it, twice.

This has been a good week for paperwork on Planet Alex. I signed the lease for my long-term apartment this week, which was a much more involved process than any other contract signing I’ve ever done, including the ones through [name of a random previous employer]’s procurement system. 10 sheets of A3 paper. My wonderful agent even had a pen and a bottle of water ready for my troubles. If anyone is looking for property in Tokyo, I will definitely connect you, please ask.

Running continues. Still playing it slow, but I have ramped up a little bit and am seeing marginal improvements in recovery. I ran between my current and future apartments which made Tokyo feel rather small. Now if only I could find a shower in my office building, I could try a commute...

When you move, in Japan, you must go to your ward office to change your address. This is basically the same process as when you register for the first time, so the process I did back in November. I did it all on my own this week, it only took about an hour, and it only involved writing a tiny bit of Japanese - which I'd learned in class!

I spent a lot of time in the AI headspace this week, partly playing catch-up with some very fast new developments.

• On conversational interfaces: The real headline, for me, is this demo by Sesame: I had a conversation in which we nearly passed the Turing test. I cannot explain what I experienced, even though the paper explaining it is right beneath, I can only recommend trying it for yourself.
• On research assistance: I tried them all, and I mean, all, including a few open-source ones. I polled a few people and groupchats to compare the outputs. OpenAI’s genuinely blows all of the others out of the water, although admittedly Google still haven’t unveiled their current-gen model. Google’s was the most ambitious and drew upon the greatest number of sources, but I found it consistently lacking, it just didn’t give me the information I wanted to know. But I'm betting big on Kagi in the long run.
• On coding assistance: Anthropic’s new coding tool is cool, sure, but, it’s just more power thrown at the same problem. These OG frontier models seem like they’ve totally stalled and are relying on social media hype to silence ongoing research and competition. What really blew me away here is Mercury which is a family of diffusion models which is the first significantly different LLM architecture I have seen. Check out how fast the demo is!
• On g3: I wanted to address this one in particular, because given recent hype I had to see for myself. I did. It made me sick to my stomach. The technical catch-up rivals DeepSeek in pace and the UX is very good given what happened to Twitter, but the model has been contorted into a deeply strange shape, one which showcases the immaturity of the people behind it. The words "unhinged" and "sexy" are thrown around, but in reality it is but the personality of contrarian tween going through an awkward puberty.
• However! All in all, the tech is very cool, and I’m quite happy to see that the new players are pushing for lower power consumption, more ethical data-sourcing and significantly more humanistic interfaces. This is great. And hopefully, if Satya turns out to be wrong about AI, it isn’t all bad news anymore.

Monday was a bank holiday so after sauntering around exploring my new hood in the morning, I dropped off the keys to my last place and I went out to Samurai Fest which was a good bit of fun.

Got some sad news on Wednesday, offset by some good news on Friday.

My commute to work is a 25 minute walk, which is still longer than my 18 minute walking commute from Sofia. On Friday I managed the reverse trip in 20 minutes, which I still don’t understand.

Why the rush on Friday? I was going to the Hacker News Tokyo meetup for an “interactive story-driven concert”, by Extra Lives who were touring in Japan. It’s hard to describe it. It was very, very, wonderfully cheesy. And a little bit cheeky, from like, a humour perspective. I can only recommend it. They’re touring London in April, and some other places, so check ‘em out.

OK but the real reason I rushed on Friday was because Nina Kraviz was in town. Kraviz has been on my DJ bucket list for just under a decade (after the one-and-only EB fed me propaganda) but I just hadn’t managed to make it work. Until Friday. Seeing her first-hand, I get it: She’s up there in the upper echelons of technical DJ ability, and I was genuinely blown away by (for example) the layering in her set.

Unfortunately my suspiciously-internationally-skewed fellow audience-members were not as blown away, nor a regular club-going crowd. Lots of pushing and shoving to try and get to the front, not to dance, but to take photographs/selfies and then to leave for the bar. It’s poor club etiquette anywhere in the world — trust me, I’ve been around — and I’m surprised to see it in Tokyo of all places. I guess the cheap yen and the promise of international acts pulls in all sorts. Tokyo is the new Sunny Beach.

For the same reason, I did not in fact see TJ last week. I got to the incredibly tiny venue, took one look at the queue, and thought better of it. :^(

I try not to think about what the weak currency is doing to my personal finances. It probably isn’t good. But after much paperwork, I now have a NISA, so, you know. The textbook says that I’m doing it right?

Tried to invest in my brother’s new start-up and discovered that the fundraising platform specifically bans people living in Japan. I already had some investments with them, and am now apparently unable to participate in ongoing fundraising rounds for those too. Thanks lxds!

I published last week late, because I struggled to articulate something, but didn't want to publish to socials. Shout out to RSS and email crew. It’s there if you want to see it.

have a great week!