This is my highest conviction problem with web3. We need to build things that hold our values native, but serve people through the unlock of new and meaningfully significant value not possible before and not in of them selves just a value we already might hold.

When mathematician’s solved problems on graphic calculators, did Casio announce the news as their success? Unless it’s a single *prompt then hands off*, how do we appropriately recognise the level of expertise needed to massage the model to get the breakthrough outcome, as to clarify for where the result is something a layperson can experience. To imply otherwise would be disingenuous.

Though I didn’t immediately appreciate it, I think there's some worthy pause needed to recognise that Spotify's new logo literally emits energy toward the user. I really don't think I've come across something like it before. (in terms of app logos) I often feel excited when returning to music after realising that whatever made me pause in the first place was now gone... but I have really felt like spotify is greeting me with shared excitement now vs the old logo. As if it's saying/agreeing with me: "That's right, it's disco-party-tune time baby!". Perhaps I'm internal-meta'ing this one a bit far, but I think it's placed fairly for me to say this is unique, compelling and perhaps indicative of what clean and minimal logs might never achieve.

I feel like this is a unique category of predictive generation error that I've not come across before. pop-hallucination?

Ngl kinda neat. Though I’ve refrained from Sora due to the account-locking-T&Cs.

Making them like 4ft tall is such a solid design choice. Hard to be afraid of little shorty-bot helping around the house like that.

If you aren’t crafting your own agents from scratch, or consciously spending the time to understand and tailor the agents you adopt from others, they aren’t *your* agents. Not inherently good or bad, but I think it’s important to be mindful of the trade-offs being made and whether the tasks you delegate are material, involve nuance or afford discretion. Likely scaled against complexity. What I mean by this is to recognise that how an agent is designed and scripted is attached to the values of its author, and this carry’s the implication that the agents you adopt will offer reasoning and guidance that (to varying degree) are skewed by the values that contributed to how go about their work and how they communicate it forward/in reply. This may lead to a different outcome than you would want if you were able to pause time if need be to go end to end yourself. side note: similarly a layer deeper, you could say the same attention should be given to the data substrate of LLM itself. Though that’s not as realistic to directly own or influence, it may help in how you think of promoting for mitigations or to see where it’s appropriate you use multiple models.

I feel for the thief who dropped the crown during the Louvre heist. You can just imagine the level of chat he'll now endure from his bros in perpetuity. Made it big, but cry-smiles watching them re-enact his fumble for the 800th time while they hangout on the yacht.

Someone should make a little independent oracle service of people who report on the queue lengths across the gold stores of major cities in different markets to index the retail sentiment. With enough you could see this index fed on-chain. Should also include local weather reports to account for wet-weather dips (or increase weight if it’s pouring and the queue remains strong). They could also survey people in line with some type of robust and thoughtfully objective/non-leading questions to add more pov and colour w/ stats on demos… etc.

Little tip I stumbled upon and haven’t seen elsewhere 🐢 -> Reduce audio/video playback speed for complex topics to 75%. The classic ‘2x’ is great for fluffy generals or podcast chat topics, but if your learning something new, slowing it down it really does seem to both (a) hit harder and (b) give space to allow for the uncontrollable pondering you’ll have without losing place and needing to skip-back to revisit what you accidentally tuned out of. I’m on the fence with 50%, haven’t tried enough to break past the novelty.

Just a feelgood share 🐠 Louis “Father Fish” Foxwell has the most wholesome (and tremendously helpful) fish keeping channel/ecosystem. Invented his own self-maintaining tank methodology over the last 70 years and launches some of the best guides I’ve come across. He runs a better managed discord than most savvy kids could do w/ 20k members and all sorts of niche q/a topic support. I don’t think of retirement often but I think he’s nailed it in the way of retire-without-stopping type of way. youtu.be/TDuF5zuCyHM?si=OaNF… @fatherfishtanks

Art

there's something kinda poetic that every illustration portraying the multiverse has infinite bubbles 🫧 Can AI really pop if evolves to take us there?

GPT-5 seems far better at unprompted lateral reasoning than earlier chatGPT models. While the cost-led performance throttling has been painful, I think the value of getting responses which add things that may be useful to consider (but weren’t asked for) is enormous.