weekend ai reads for 2025-12-05

2. An early version of DeepSeek, the Chinese AI, “did exceptionally well at reasoning, but had the crippling weakness that its outputs were an unintelligible mix of Chinese and English.”

13. That cardamom, it is cheap but tastes expensive, if cardamom cost 1000× more, people would brag about how they flew to Sri Lanka so they could taste chai made with fresh cardamom and swear that it changed their whole life.

5. But nobody talks about the Canadian Invasion of the 2010s. Drake. The Weekend. Justin Bieber. Carly Rae Jepsen. Shawn Mendes. Even that horrible one-hit wonder Magic!

i didn't think he ever had a chance to be happy or authentic, his entire life. he sacrificed himself so i could.

To them I’m a silly little idiot because that’s what they desire me to be. I guess sometimes that’s just part of the deal.

What made the difference in my year was having a system connected to an astonishing breadth of knowledge that would stay with me, with patience, through every messy step of running a one person company. ChatGPT Pro has provided me unlimited patience. I have asked questions that experienced developers might laugh at. It answered them straightforwardly, over and over, on its thousandth repetition as calmly as on its first. It did not care whether I was supposedly vibe coding. It cared whether the code compiled and whether the system worked for my goals.

This does not mean AI Safety is 100% solved yet — we still need to triple-check these proposals, and get engineers/policymakers to even know about these solutions, let alone implement them. But for now, I’d say: “lots of work to do, but lots of promising starts”!

You can use a lot of AI technology as it relates to performance. What you’ll see in game planning as we move forward is coaches using AI to break down film or break down tendencies. I’m sure we can plug in different formulas now and say, What’s the tendency of Texas’ offense based on one game? So you’re going to see some of that. I don’t think anyone has captured the market yet on what it truly means for college sports. We’re all going to evolve.

If you’re building AI workflows inside a large org, TELUS has some advice:

- Instead of getting locked into one model, use a flexible, multi-LLM platform that can adapt to new tools.

- Think like a product team – design a system that people will want to use.

- Use only straightforward interfaces. A good copilot builder should be as plug and play as a Canva template.

- Pair flexibility with strong governance. TELUS’s control plane ensures security and compliance without slowing innovation down.

What if education worked the same way? Not ‘don’t use AI’ or ‘use AI freely,’ but ‘show me your navigation.’ Document your prompts that led nowhere. Explain why you rejected certain suggestions. Share the moment when something clicked, even if you can’t fully say why. When did you explore freely? When did you delegate? When did you restrict yourself? When did you work without it entirely? The skills are different.

Karpathy’s core argument is simple: educators can’t put the technology back in the bottle. He says schools should assume that any work done outside the classroom has used AI.

Technical detection tools won’t fix that. “You will never be able to detect the use of AI in homework. Full stop,” Karpathy wrote on [Twitter].

Parental demand rises with the share of other teenagers using the technology, with social forces increasing willingness to pay for AI by more than 60%. Providing information about potentially adverse effects of unstructured AI use negatively shifts beliefs about the merits of AI, but does not change individual demand. Instead, this information increases parents’ preference for banning AI in schools.

1) New things to know

2) New things to do

3) Better ways to show

In other words, LLMs are actually pretty good at inferring what a user is doing from basic audio and motion signals, even when they’re not specifically trained for that. Moreover, when given just a single example, their accuracy improves even further.

This is a search tool that will only return content created before ChatGPT’s first public release on November 30, 2022.

Creating a pattern requires considering the entire work; each step has to fit with and work with all the others. Blindly selecting the next step without that consideration will, more often than not, fail. This is especially true since AI can’t “test” the pattern after writing it, which is a big part of what humans do.

As such, AI is prone to hallucinating patterns that don’t work, are physically impossible, or don’t produce what is advertised. However, this hasn’t stopped shady sellers from turning to various sites to sell AI-generated patterns.

Does the leadership at my company promote a xenophobic agenda and use the wealth I help them acquire to donate directly to bigoted causes and politicians I find despicable? Yeah, sure. Did I celebrate my last birthday at Drag Brunch? Also yes. I even tipped with five-dollar bills. I contain multitudes, and would appreciate it if you focused on the brunch one.