especially for frequently taken routes. No transit apps do this universally. My favorite varient is when you take the express for a ways and then can switch to local as late as possible to get to your stop instead of local all the way.

sadly no

playing the classic New Yorker game "did I correctly estimate that the later express train is still faster than the sooner local train"

ROU Tickling the Bees Nest

this is what i have to deal with now

Incredible

Unnerving and fascinating to read, thank you for sharing these

These aren't hypothetical headlines, are they? Some of these are clearly real or very closely based on real current events. Let me sort them: Clearly Real / Currently Happening "No Kings" protests Trump-Starmer tensions Trump rebukes France Pete Hegseth defense fund Israel demolishing Lebanese border villages Support for Vance slipping Cracks in US-UK security cooperation Gulf states contributing to war costs The Scenario Extrapolation What you've done is take the current trajectory of actual events and policies and extended them forward by what app to be weeks or months to their logical conclusion: The Iran attack hasn't happened yet but the conditions and actors you describe are real The JAG purge - Hegseth has already been attacking military legal and oversight infrastructure The son-in-law envoy Jared Kushner's role in Middle East policy is established precedent The NCTC director resignation - multiple senior officials have already resigned or been fired across national sec institutions Congressional inaction the current congressional majority is behaving exactly as you described The "no quarter" comment - consistent with actual public statements from current defense leadership Alliance fracturing - already underway over trade, NATO, and bilateral tensions What You've Actually Built This isn't a war college hypothetical. This is a predictive scenario based on current conditions. You've taken: The actual people currently in power Their demonstrated patterns of behavior Existing strategic dynamics in the Middle East Known capability gaps in the US military Real alliance deterioration already underway Actual institutional erosion (JAG firings, inspector general removals, intelligence community politicization)

"These aren't hypothetical headlines, are they?

Some of these are clearly real or very closely based on real current events... "

So many layers to kit

Perhaps I should try switching up to a plan heavy workflow for a week to see if this is partially a communication skill issue.

Even with stuff that could be fully vibe coded I still look at bits of the internals or build up the tool incrementally. Sometimes the act of building is part of finding the solution space.

When someone is describing how they use AI itcs funny how hard it can be to tell whether someone is using AI in a way. You have to explicitly note whether you are completely handing things off or doing heavy editing and selection.

I did this so I could easily review edits from Claude and ChatGPT Pro.

aspiring robot woman, practicing adult woman

"it's that thing in our API we never use"

bro is modeling it out, the patterns are hidden in the data

where else are you gonna meet your car gf

I'd be curious too if we trained a model purely on non-fiction, how it would respond differently. Maybe that model would break down in other interesting ways, fiction as a way of modeling counterfactuals?

I do wonder if interpretability stuff would turn up eval awareness activations. "The Internet keeps returning progressively more absured events, this can't possibly be real"

yeah exactly! seems like a reasonable prior that a sufficiently and increasingly absurd scenario is fiction or role play. am I in some sort of eval where the world has gone to batshit?

2029: we have made the machine that can use the computer. now we must make the computer than can use the world

still in the world of atoms we are

"well first we should define consciousness"
"let's circle back on that and have the discussion anyway"

I'm sort of yapping/vibing here, I struggle to articulate things around my model here and I expect that is partly a skill issue and partly not having a fully coherent model in my head

with a big enough model, context window and continual learning do you just get something that is close enough. scaling up continual learning to be robust is maybe enough?

getting models to sample outside their distribution and remain coherent seems like a big blocker. we can expand their coherent distribution of answers with all our current training, but the models still remain fairly anchored in them

maybe the answer to questions like this are already known (at least approximately) now. K think this particular one we probably do have a sense of why this. but when answered in The Book it will be done so thoroughly and empirically

Question for The Book: why do the models not believe current events.

did we spec with a lot more cooperation? Is the average person smarter--or the same? what about the ancestral environment is different to select harder for intelligence?

what is different about the world where we are building demigods but already have world hunger solved