@odbol@giansegato Jumpcut hasn't had a commit in 3 years, not sure they're taking PRs lol. It also doesn't have a CLI or search API like clipmem, and uses about 2x the memory for 10x slower polling interval while being 125 ms slower during a normal copy. Plus, it's fun to build your own stuff!
Running my updated apocalypse-bench questions against a bunch of small models to see which you'd rather have saved on your laptop in the event of an apocalypse.
Tests medicine, engineering, materials, chemistry, and many more categories.
As you'd expect the largest densest model, Gemma 4 31B, is currently winning. Early days though, maybe Nemotron can take the lead.
spent my 11-hour flight back from europe working on a very long report. started as a slack message but morphed into a several pages long doc. wifi was as shitty as it gets. after finally making it home i realized that the computer had forcefully restarted. opened slack: draft was gone :(
hail mary: claude pls save me, no clue how but pls try
it checked APFS snapshots, time machine, slack indexeddb, write-ahead logs, service worker / http caches, local storage, app logs, hibernation image... nothing. all gone
but then... it realized i have alfred installed. so it checked the clipboard snapshots alfred keeps in sqlite. sad news: alfred clipboard memory gets deleted after 24h. aggressive retention policy. however! when sqlite runs DELETE, nothing gets actually deleted. it only marks pages as reusable, but it doesn't override the physical bytes. so claude decided to do a raw-scan of the db, reverse eng alfred data format, figure out the portion containing the timestamp, stitched everything back together across overflow pages... and handed me the exact final version of my report, the last one i cmd+C'd
all this, in a single shot
... day 200 of "what if you had an elite hacker you can ask anything to"
@theo@amar_patel@Teknium When I discovered Hermes was just made by crypto bros all the vitriol the creator has against any criticism (or god forbid any comparison in OpenClaw's favour) made way more sense
Decided to try a /goal with Opus 4.8 overnight and it just absolutely destroyed my codebase. Codex had to clean it up. How does anyone do anything serious with Claude models?
This is a dumb person's idea of how a smart person thinks. The answer is obviously, *obviously*, 72.
If you sit there autistically and complain that you don't have all the required information, then you haven't understood the question, you can't make reasonable assumptions, you can't do basic inference, and you're not a problem solver.
This is why IQ tests don't work too well on really smart people.
Because sorta smart people tend to give the expected answer.
And really smart people tend to point out that the question is wrong, and start arguing with the test, or trying to correct it, thereby making the test impossible to grade and annoying everyone.
The expected answer to this is 72.
Because 2*2*2 = 8 and 5*5*2 = 50, so 6*6*2 = 72.
But the (really) correct answer is "I don't know."
Because what you have is two points on a 3 dimensional graph (x,y) -> z.
z = 2*x*y is one surface that can be drawn through these two points. And I suspect it's the simplest formula for a surface that can be so drawn, although I haven't bothered to check.
But an infinite number of contiguous surfaces can be drawn in three dimensions that encompass these points (2,2,8) and (5,5,50).
Each of these surfaces can be described by its own formula. Some of them will also touch (6,6,72). But others of them will touch (6,6, {something else entirely}) instead.
This might sound really, really pedantic. But it's not.
Everyone knows that the expected answer is the simple one, but that's only on a test... a fake artificial made up problem.
When we start trying to do this in the real world, which, after all is what this "IQ" thing is actually for, then using the same kind of "IQ test thinking" can get you in trouble.
"My 3-month-old son is now TWICE as big as when he was born. He's on track to weigh 7.5 trillion pounds by age 10." -@pronounced_kyle
Fitting the simplest-formula curve, as opposed to the correct curve, makes our predictions of real-world stuff dead wrong.
So this kind of test question promotes a dangerous habit of thought.
But, Devon, I hear some of you ask, doesn't the principle of Occam's Razor demand that we fit the simplest curve?
No. No, it does not. It does not require that we select the simplest possible answer, given what we have currently seen. It requires that we prefer hypotheses that make fewer assumption to those that make more.
These are two different things entirely.
If I see one black sheep, the simplest hypothesis is that all sheep are black.
The hypothesis requiring the fewest assumptions is that at least one sheep is black on at least one side.
You will note which of these is correct.
All of this is, of course, irrelevant to questions on IQ test. But questions on an IQ test only matter as much as they are relevant to the actual universe...
Where ideas like this are very relevant indeed.
sifs does symbol lookup and file outlines now. cached navigation metadata means agents don't rebuild the full index for structural queries. outlines come back in under a millisecond.
wins 80% of head-to-head tasks against codedb. symbols at 0.78x their latency, outlines at 0.57x.
@HeiligGerh36381@itsolelehmann No you won't. You'll feel better at a comfortable temperature. If it's 35°C outside there are no clever tricks to make it feel 20 inside. You need a heat pump to pump heat out of the building.