> More recently, media-streaming devices have been identified as a major carrier of malicious scraping software. Sometimes the devices are compromised at the source; other times, they are just poorly secured and easily compromised after the fact.
I run an OPNsense firewall at home and the OpenWRT router at a hackerspace. Are there ways of auditing that devices aren't compromised? Tracking which devices still send lots of data when no one else is using the network?
One-shot benchmarks are great for me as a solo creator, since they slightly correlate to whether the better frontier models (Opus and Fable for me) make better decisions about things I didn't spec, or whether they'll give me better suggestions right off the bat.
I think at some point, AI development becomes a competition for who has the best taste [1], because with seemingly unlimited/cheap tokens, AI ends up being more about throwing shit at a wall and hoping something sticks.
> Coca-Cola doesn't "quietly water down" its product to save a few bucks. They know people will take a sip, say "oh that's not what i wanted", and go buy a Pepsi.
Coca-Cola is also mostly measurable and reverse-engineerable.
The Claude models are black boxes, and actively curtail distilling efforts.
I've been using Compound Engineering for the past few months too, but now that I'm seeing mention of how Superpowers consumes a lot of tokens and context, I wonder if Compound Engineering does the same.
I often end my Claude Opus 1M sessions at around 60-80% context, and that's with doing `/context` once in a while, and forcing the agent to wrap up and write handover notes, so that I can start on a new unit with fresh context.
They also have the superpower of having unlimited tokens and effectively unmetered budgets (I remember a thread discussing about Steipete spending $1.3M in tokens in a month [1]), and being able to practically bruteforce anything to life.
This comment reminded me of the James Hoffmann video on the Philips Baristina coffee machine [1].
Philips is in no way strapped of engineering talent, but they can still design the occasional bad product, when there are many alternatives at similar price points that are better, like the Sage Barista Express.
On the other hand, if they'd focus on building their core product (an oven) while taking BigCompany's money, and without listening to the multitude of smaller requests from potential customers who might not even hand over a single cent, they might have eventually gotten somewhere.
I agree that being a "custom engineering contractor for BigCustomer" might not have been the ideal outcome, but at least the entire company would have gotten some expertise, and probably be able to pivot and finally build their own thing given more resources (*cough*moreVCfunding*cough*).
> We had this at my old work, except it wasn't a display, it was a circular piece of paper with "clean" on one side and "dirty" on the other. When it was done, you rotate the paper so it was clean side up.
The downside of low tech solutions if many people touch it (i.e. not just the janitor or rostered staff) is that someone can still add their shared mug to the dishwasher just after a cycle's complete, causing the entire batch to be considered unclean. Or in your example, someone deciding to flip the sign but not emptying out the dishwasher.
Then again, these are little things that people have learnt to make do and just live with, and I'm not sure it's worth paying 2x or 3x, or even 20x the price just for fancy feet, like what the article thinks people might do with a fancy oven.
> This prompt reveals how absolutely _ass_ the Claude Code harness is (so many stupid tool call failures)
I've just started using Claude Code this month after months of Claude in VSCode + GitHub Copilot (and a bit of dabbling with AWS Kiro), and I'm actually impressed by how seemingly polished Claude Code is.
I think Copilot in VSCode broke far more in my months of (ab)using it.
The thing that stood out for me about Meituan was that their power bank rental gizmos were everywhere in China, and people would rather rent a power bank than own and carry one around because of how convenient it is.
Speaking of images (media file), I found out that iCloud double-counts storage space for rotated photos, so I need to subscribe to a higher tier just because the iPhone often can't get orientation right, especially for food photos.
Would be nice if they moved to storing diff sidecars.
> But then I've got a whitelist of allowed domains, which I updated on-the-go if that one site wife really needed wasn't working due to overzealous filtering.
That's fine. I'm just puzzled why people would want to use a "block ads and tracking" DNS server without a whitelist feature.