Man, Machine, Self

Twitter data science project that probably can't exist:

survivorship curve of all accounts who ever posted "never deleting this app".

Solution to this benchmark, as with many things in life, is mercilessly stealing plot hooks from old Encyclopedia Brown stories

Scroll till I find stuff I wanna boost or shout at, then do that for a bit, then milk some dopamine from numeric badges appearing, then repeat when I don’t have enough badges to hit my red queen hedonic baseline

To watch for this, listen if the commercial OS pivot sales pitch to normies becomes

“your computer has a ‘car key’ now which we bundle with purchase. Unless it’s in, assume agents and adversaries will be blowing up your machine!”

Security people will (through gritted teeth) support the noble lie.

I also think mass market physicalized auth is also going to get its shot at the mike, because root access control without your yubi can be sold as

“oh without it, the assistant your computer WILL eventually destroy your machine!”

They’ll be selling the fix to the problem they made, but hey.

Much more subcomputing. Code lives here, compute lives here, network access lives here, this is your “drive” just for code, this is your “drive” for the browser”, and so on.

Limiting fire radius WHEN zero days hit will be the focus.

This will be paired with top focus on internal anomalistic behavior detection:

focus will shift to catching and isolating out of distribution nodes proactively, and the ability to quickly burn down and isolate components of a mostly dumb terminal interfacing between monitored subcomponents.

Focus will correspondingly shift from hygiene and resource gating (is this rate limited, is this vulnerable to this, or that attack, etc.)

to risk management via interoperability: assume a zero days in [component], how quickly can we mechanically divest or switch off it?

My expectation is “true zero days will be massively scarier due to ease of scaling, but most typical security flaws will be massively less common.”

Net result will be security moving to a “black swan” pattern where script kiddie stuff basically doesn’t work anymore but threat actors are terrifying.

After just twenty minutes of exploration, Claude Opus 4.6 reported that it had identified a Use After Free (a type of memory vulnerability that could allow attackers to overwrite data with arbitrary malicious content) in the JavaScript engine. One of our researchers validated this bug in an independent virtual machine with the latest Firefox release, then forwarded it to two other Anthropic researchers, who also validated the bug. We then filed a bug report in Bugzilla, Mozilla’s issue tracker, along with a description of the vulnerability and a proposed patch (written by Claude and validated by the reporting team) to help triage the root cause. In the time it took us to validate and submit this first vulnerability to Firefox, Claude had already discovered fifty more unique crashing inputs. While we were triaging these crashes, a researcher from Mozilla reached out to us. After a technical discussion about our respective processes and sharing a few more vulnerabilities we had manually validated, they encouraged us to submit all of our findings in bulk without validating each one, even if we weren’t confident that all of the crashing test cases had security implications. By the end of this effort, we had scanned nearly 6,000 C++ files and submitted a total of 112 unique reports, including the high- and moderate-severity vulnerabilities mentioned above. Most issues have been fixed in Firefox 148, with the remainder to be fixed in upcoming releases.

this is wild lol

Absolutely cursed
Well done

Frumious jestermaxx

All ethical systems morally mandate an instantaneous “nah just your mom” in reply

Jupyter walked so my kludgepile of dockerized task specific react dashboards could run

Mine's changed like six times and is presently a weird interconnected web of folders, but the high level is

"most of the value is telling your agents how you want them to approach tasks, and principally any super-structure SHOULD work as long as you and the agent are both aware it is to be used."

The societal consequences of “recording technology transformed poetry from “THE premier wordcel flex”

to “the market for lemons on musical talent” are still reverberating

Bruh honestly I’ve received the exact same letter at least like 20 times in my corporate life from security departments

The machine’s just like me frfr

Often, but usually couched in "it's substituting for thinking in the coding process", as shibboleth for "LLM code isn't real code" or "people using LLMs are not real programmers".

Position's been getting notably quieter as AI improved & most programmers found some place for it in their workflows.

Unsatisfying handwavey answer

“I suspect they do, as the on-the-ground conditions are the same, BUT one or more of
-company focus or
-B2B target customers or
-data focus

bake a much deeper ‘don’t be weird + don’t make it weird’ filter somewhere into Gemini and Claude that OAI misses.

Scott in blogpost's on that

"Fezzik from the Princess Bride telling you "I just want you to feel you are doing well", rather than just braining you with the boulder he trivially can throw at scene start"

-strat. Definitely doesn't win as much, but does make more friends.

They hit the "no yappin" button and I Will Never Recover

Tentative candidate list:

"could of"
oxford commas
wrong less/fewer
wrong two/too/to
wrong then/than
wrong your/you're
wrong affect/effect
wrong who/whom
wrong its/it's
"a women"
Sentence ends w/ a preposition
Comma splice
greengrocers apostrophe's
"me and him went to..."
run-on sentences

Latent space tier list: "grammar errors" to "wordcel aura loss"

Top tier "absolute, immediate, permanent".
Bottom-tier "useless info-hazard dodged, aura gained".

Altogether, quite cozy. Happy to be here.

New posts get handled better on bsky and it’s NOT close.

bsky much more heavily favors “stuff your network likes”, and if you have good people in that network, it’s both a firehose of bangers, and your posts are far more likely to get seen by a non-random, non-bot subset.

VERY welcome change.

Re: Iran war stuff, bsky and X feel tied on “speed”, X and bsky picked it up around the same time.

X not having a speed premium is a pleasing surprise, and a welcome one.

- Other gap, BUT crucially not loss, was inline translation. Had a brief conversation with some Finns here, and google translate round trips are inferior to inline “just write your language”.

Not a loss by bsky imo b/c X never felt “merged” enough so cross-language community abilities would matter.

Decamped to bluesky + unsubbed from x day 1 feelings

- it’s wild how the winds have changed on usage: my xposts here dramatically out-engage my OPs

- I do miss longposts on X, and here, this is THE big feature-gap for me. Pretty much everywhere else besides this and feed velocity, bsky wins/ties.

I'm the person who wrote the tweet above.

I wrote four blog posts condemning Elon killing the children, argued about it on Twitter extensively, and donated $50,000 (partly my own money, partly contributed by other EAs who read my blog) to help the African clinics effected.