🔥🔥🔥

I realise Deepseek is very two weeks ago, but I was trying out a new extraction prompt on its reasoning model and umm….

💯

This has such a weird defensive energy delivered via a jumble of ideas.

The real kicker is at the end on why China shouldn’t get chips: “they're beholden to an authoritarian government that has committed human rights violations, has behaved aggressively on the world stage”

Bro, I have some news 👀

Buttondown, Beehiiv, Ghost.
Congrats on not supporting nazis. ✌️

“It’s strange having to go around and do eight laps.” 🔁

Australia’s Cam Myers shares his first impressions of indoor track after running 3:53.12 for a new U20 indoor mile world record in his first race under a roof.

🎥 INTERVIEW: youtu.be/PvUl9YNgM1o?...

🔥🔥🔥 Canberra’s finest 🔥🔥🔥

This is amazing. Cannot wait to listen my 5 favourite songs and top friends and put a rainbow gif as my background.
skyspace.me/index.html?n...

Donate to Wikipedia.

Do it today, and every time one of the broligarchs has a tantrum because he can’t buy it

Side note - I'd be interested in replicability here. Is the model still essentially taking a random walk or is there a level of quality (or more realistically higher weights) that consistently gets to better and/or consistent outcomes through reasoning.

This doesn't seem to be what's now going on in your example. It actually does seem to be plotting its own path forward to a response. Is there prompting (or an initial instruction) that is hidden and directing this reasoning? Or is it baked into the model somehow?

I haven't been reading a lot lately on this, but ~12m ago it seemed asking an LLM to reason post-response would be convincing but also obviously not actually the reasoning that led to the initial response.
A mental model was "here is a point at the base of a hill, now make up your path from the top"

New season of The Man in the High Castle is weird.

Doing god’s work.

I knew one day I’d have to watch powerful men burn the world down – I just didn’t expect them to be such losers

Can a headline alone be eligible for a Pulitzer?

A handwritten note that is practically indecipherable apart from the words “TREE LOPPING”.

Unexpected use case for Claude this week - reading notes from my neighbour. He’s in his 80s and organising trees on our border to be trimmed. Absolutely no chance I can read his handwriting so uploaded to Claude which (somewhat surprisingly) nailed it.

No I definitely read the post. (Maybe don’t jump to conclusions?) The author is reporting on people not using AI for real-time information on an emergency. He then links that to a lack of trust in AI. He should be linking it to a lack of suitability of AI tools for that task.

Another banger in the genre of hot takes from substackers who are confused about what AI is actually useful for and therefore concluding it is dumb.

‘The pandemic isn’t over’: my year of long Covid – video Darren Parkinson is one of about 2 million people living with long Covid in England and Scotland. The illness is having a detrimental impact on his life, stopping him from being the kind of active and...

The Guardian have documented a year in my life living with #LongCovid.

Please watch this short documentary to learn what it's like living with this debilitating disease and why we need urgent treatments.

www.theguardian.com/society/vide...

Perhaps they fingered you?

Telstra and Accenture agree $700m global AI joint venture Telstra and Accenture have agreed to a proposed joint venture to accelerate Telstra's data and artificial intelligence roadmap.

Telstra bringing in Accenture to "do AI" at a price tag of $100m per year. So they expect $100m annualised cost out at a minimum. Retail customers will likely never be able to talk to a human at Telstra again.
www.capitalbrief.com/briefing/tel...

From the DJs community on Reddit: Soundcloud April Fools joke? "Here's the Drop™" Explore this post and more from the DJs community

Just reminded me of a rare good tech April fools, when Soundcloud “introduced” the new “feature” which was a big arrow saying “Here’s the Drop”.
www.reddit.com/r/DJs/commen...

I’ve been enjoying your veo experiments. I’m interested (and can’t really come up with a good answer myself) on how useful these text to video models are. Besides playing around with them, where do you see the value in the long term? (that justifies their expense to both build and run)

Australian COVID-19 weekly stats update:

The risk estimate has risen to 1.2% Currently Infectious, or 1-in-86. That implies a 30% chance that there is someone infectious in a group of 30.

#COVID19 #Australia
🧵

25mm of rain in 20 minutes resulted in a skylight leak over the box sets. Drying them of and it strikes me that back in the day they were all-in on terrifying covers.
(Brahms Requiem I can understand, but why make Casals look like a devil? And the Beethoven sonatas are lovely)

after days of grifters and con men at CES we stumbled upon the booth for VLC. they were all dressed as wizards and told us, "we have nothing to sell, we just decided to show up". i told them I'd been using their software to pirate media for 15 years and they said "keep doing that"

Increasingly convinced that folks like Zuckerberg, Musk, Bezos, Andreessen and those All-In podcast clowns spent their teens desperately wanting to be @mods on IRC channels and now we’re finding out why nobody let them.

If you’re outraged by this I have some news for you about the business model of Facebook, Instagram, TikTok and Google. 🤨

On this day 12 years ago, @kcg.bsky.social posted the “This is Fine” comic.