Kohan Ikin

Another +1 to Anthropic posting here - and that ad campaign was great too. I'm starting to get Apple Mac 1984 vibes from some of their new ads, in an inspiring way.

YouTube video by Alberta Tech Vibe coding fixed my impostor syndrome

I know I'm more than a week late to this one, but this was one hell of a mic drop and absolutely worth spending 1 minute on programming humor:

YouTube video by Rick Beato This Record Label Is Trying To SILENCE Me

Universal Music Group has given Rick Beato 4 days to delete his interviews with Universal artists, or his entire channel will be deleted from YouTube:

www.youtube.com/watch?v=zBq_...

Live: Sweden's historic 600-ton Kiruna Church begins move to new home YouTube video by The Independent

SVT has a live broadcast of the moving of Kiruna's church to its new location!

Kiruna is the Swedish town where all the buildings are being moved 2 miles east, due to subsidence caused by a nearby mine.

I've had at least two occasions where it felt like Sonnet advertised to me. Seems that if you ask a prompt with a clear near-term purchase intention, there's some kind of mode that can occur. Claude insists they aren't ads though, and the choices were aligned with what I wanted.

Eurovision Song Contest gets brand refresh to celebrate 70 years of world’s largest live music event From today, fans will see the Eurovision Song Contest is wearing a new look. The Eurovision Song Contest is celebrating seven decades of being United by Music in 2026 and marking the event in true Eur...

New branding and a logo refresh for Eurovision. Apparently the new font is called Singing Sans.

I just *had* to test it, the link is a 404. They could at least have added a redirect!

That new Gemma3 270M AI model from Google really is cute! I tried asking it that same question "What is it like to be a bat?" and I got back a cute tool call:

{"name": "bat", "arguments": {"action":"play"} }

I think Gemma3 was disappointed there was no bat to actually play with 😥🦇

The redesigned Pebble Time 2, a stainless steel model. The new model is much more curved on all edges, with 3 distinct hardware buttons on the right and one button on the left (not visible). The watch has a standard black wristband, but will take other 22mm bands, including the original Time Steel wristband. It looks awesome.

The new hardware redesign for the Pebble Time 2 looks fantastic... okay, count me in, pre-ordered. Looking forward to 30 days battery life & heart-rate monitor too. @ericmigi.com thank you *so much* for bringing the Time 2 design closer to the Pebble Time Steel aesthetic!

YouTube video by Kai Lentit Adventures Don’t go clubbing East Berlin this summer.

If you like / miss Berlin, or you just like German humor, this Kai Lentit video on attempting to go clubbing is hilarious.

I'm going to have to go back to Berlin one day, aren't I?

Like half the world, I'm busy downloading the new open source OpenAI model onto my computer now. 14GB just as I go to sleep, hopefully will be ready by when I wake up.

Also hoping gpt-oss:20b will somehow run on my 16GB RAM machine, no matter how slowly. Would like to have an offline alternative.

America needs to take open models more seriously. Just this summer the early lead in open model adoption of Llama & co has been overtaken by Chinese models.

With The American Truly Open Models (ATOM) Project we're looking to build support and express the urgency of this issue.

Claude is now successfully adding tasks to my Toodledo to-do list. It's both weird and very cool to refresh the to-do list on my phone and see tasks written & added by an AI there for me.

Also weird but cool: Claude *really* wants me to talk about The Witness game with it. High priority, due today.

Last local AI experiment of the night, downloaded Gemma 3 4B. And that model was actually useful for answering some simple code questions. Worked fine on my 16GB non-GPU machine, even if obviously not as fast as Claude.

Finally finished downloading Qwen2.5-Coder 7B (over this 4G connection because I *still* don't have internet). Tried the same prompt. Slow response & not quite the right solution, but a plausible answer. Seems I can run a 7B model without maxing out my 16GB RAM too.

Tried the same prompt with Gemma 3 1B. On the bright side, Gemma 3 answered, and super fast! Even on my Intel machine with no GPU!

Down side, Gemma 3's "solution" to this coding problem is absolute garbage. But I'm probably expecting too much from a tiny 1B model.

Internet is down, but last night I'd downloaded Ollama and a couple of AI models, so I could use AI offline.

My first try is not going well. Qwen 3 4B is stuck in an infinite thinking loop. Pages and pages of the same few sentences repeated. "But why would the user want that?" "But how to do that?"

here’s a curse word free version of “look it up” as requested by countless teachers and librarians 🥰😘

The old timers who built the early web are coding with AI like it's 1995. | Christina Wodtke The old timers who built the early web are coding with AI like it's 1995. Think about it: They gave blockchain the sniff test and walked away. Ignored crypto (and yeah, we're not rich now). NFTs got...

This comment by Christina Wodtke is spot on.

"The old timers who built the early web are coding with AI like it's 1995 again... the same folks who hand-coded HTML... When the folks who've been through every tech cycle since gopher start acting like excited newbies again, that tells you something."

The best available open weight LLMs now come from China Something that has become undeniable this month is that the best available open weight models now come from the Chinese AI labs. I continue to have a lot of love …

July has been a truly incredible month for model releases from China - Moonshot (Kimi K2), Z ai (GLM-4.5) and 5 new releases from Qwen

I think it's undeniable that the best available open weight models now come from the Chinese AI labs simonwillison.net/2025/Jul/30/...

From the ClaudeAI community on Reddit Explore this post and more from the ClaudeAI community

If you're wondering why your Claude Desktop chats are out of order with new timestamps today, you're not the only one. Seems timestamps have been updated on old Claude Opus 4 chats for some reason (new model drop soon?)

No chat history actually lost for me, they're just out of order.

Starbucks Abandons Mobile Order, Pickup-Only Stores The world’s biggest coffee chain is moving away from operating stores that only accept mobile orders for pickup because they don’t have the right vibes.

I like that Starbucks are closing their mobile-only stores as "too transactional" and "lacking warmth and human connection". Business focus is back on coffeehouses.

The product isn't the coffee (let's be honest, it's not great). The product is the ambience. Aromas, music, comfort, lighting, etc.

YouTube video by Sophia Ziesch Why I Distanced Myself From the Berlin Club Scene 🎛️

Sophia Ziesch on leaving the Berlin club scene. Hard to decide how much of this is just getting older / "back in my day", but when scenes feel unsafe, it's probably time to move on. I'm sad if that's what Berlin is like now.

youtu.be/PtfUlCBpiQk?...

Humans not quite obsolete. Struggling to compile a Go program Claude sketched out, every compile ended with "no packages to build". Claude had no ideas. Lots of debugging, none helpful.

Yeah, don't call your Go file "crosscompile_test.go". I had to remind Claude that _test.go has special meaning.

Borrowing an idea from patio11: I'm adding some English readable notes to my website's robots.txt, specifically for AIs (and specifically Claude). Not to bias or trick the AI, but suggesting which parts of pages I think would be most helpful and timeless, for the humans they are assisting.

YouTube video by Aberrant DSP Introducing Lair

A year later, and the video for Aberrant DSP's "Lair" is still one of my favorite VST plugin product videos. They absolutely knew their audience - visually, sonically, right down to the typography and descriptions. Perfectly gothic.

"There is only instability and dissolution this far out."

I'm considering ranking web pages by the percentage of actual content compared to code. I'm finding pages where the actual text is only 1.4% of all the HTML code being sent.

If you can handcode your website, and it isn't bloated with ads & trackers, that already says something about its quality.

I'm not yet convinced that using TF-IDF to choose keywords for web pages I index in my search engine has actually helped. I should treat each web domain as its own corpus with its own TF-IDF scoring, plus another corpus for the internet, and this is feeling overly complicated and my head hurts.

@brianjin.bsky.social Hey Brian! Tried visiting your website tonight (the thebrianjin.com one), but looks like the SSL certificate expired on 8th July and so Firefox won't connect to it. Might need a quick certificate update. Hope that's helpful, don't want you missing out on listeners!

My personal search engine project has somehow expanded to the point of trying to keep a local copy of "the internet" on my internal laptop SSD.

This seems both a good idea (I can add TF-IDF keyword indexing! It's like a personal Pocket / Read It Later replacement!) and a *terrible* idea.