conputer dipshit
LightNews LightNews
banner
davidcrespo.bsky.social
conputer dipshit
@davidcrespo.bsky.social
2.8K followers 570 following 7.5K posts
web dev + hot dad. enjoy charts, unions, conputer games, philosophy. chicago crespo.business
Posts Replies Media Videos
davidcrespo.bsky.social
everyone suggesting they check the quotes by hand
February 14, 2026 at 7:00 PM
davidcrespo.bsky.social
nobody's doing it with jj, you say...
⏺ Done. The landscape is remarkably homogeneous — tmux + git worktrees + bubbletea everywhere. The jj-native angle is genuinely unoccupied. Workmux's "the multiplexer IS the interface" philosophy is the closest match to the zellij prototype approach, and its .workmux.yaml for symlinking shared files across worktrees is a practical detail worth borrowing since jjw already handles the .claude/settings.local.json symlink. ✻ Worked for 4m 27s ❯ look one more time for a tui for this based around jj ⏺ Task(Find jj-based agent TUIs) ⎿  Done (51 tool uses · 42.1k tokens · 11m 1s) (ctrl+o to expand) ⏺ Nothing. The space is entirely git worktree-based. The closest thing was a claude-squad issue where someone offered to upstream jj support — it was closed without being merged. There's also jjagent (archived, by schpet) which tracked Claude sessions as jj changes via commit trailers, but it was single-session and tightly coupled to Claude Code. Plenty of blog posts describe the manual workflow (jj workspaces + parallel Claude sessions) but nobody has built the tool. Open field.
February 14, 2026 at 4:05 PM
davidcrespo.bsky.social
TIL this setting on GitHub. kinda mad about the years of git commits could have contained my beautiful PR descriptions
setting on github to set commit message of a squashed PR to PR title + description
February 13, 2026 at 11:42 PM
davidcrespo.bsky.social
my least favorite Amodei line. this is, let's say, not obviously true
www.nytimes.com/2026/02/12/o...
The United States being technologically and militarily ahead has meant that we have throw-weight around the world, augmented by our alliances with other democracies. And we’ve been able to shape a world that I think is better than the world would be if it were shaped by Russia or by China or by other authoritarian countries. And so, can we use our lead in A.I. to shape liberty around the world? There’s obviously a lot of debates about how interventionist we should be and how we should wield that power, but I’ve often worried that today, through social media, authoritarians are kind of undermining us.
February 13, 2026 at 3:01 PM
davidcrespo.bsky.social
thanks to ed for bringing this to my attention
February 13, 2026 at 2:12 PM
davidcrespo.bsky.social
this headline and the replies calling him an idiot for it are so funny:

a) DOUTHAT BRINGS IT UP
b) the centaur thing is not his idea, it's a cliche
c) critics like Cory Doctorow use this analogy all the time pluralistic.net/2025/12/05/p...
Douthat: This is where the term “centaur” gets used, right? Amodei: Yes, yes, yes. Douthat: To describe, essentially, man and horse fused — A.I. and engineer — working together. Amodei: Yeah, this is like “centaur chess.” So after Garry Kasparov was beaten by Deep Blue, there was an era that, I think, for chess was 15 or 20 years long, where a human checking the output of the A.I. playing chess was able to defeat any human or any A.I. system alone. That era at some point ended recently —— Douthat: And then it’s just the A.I. —— Amodei: And then it’s just the machine. So my worry, of course, is about that last phase. I think we’re already in our centaur phase for software. And during that centaur phase, if anything, the demand for software engineers may go up, but the period may be very brief.
February 13, 2026 at 2:10 PM
davidcrespo.bsky.social
not bad
Scenario tsc tsgo Speedup Cold (no incremental cache) 5.46s 1.54s 3.5x Incremental (one file touched) 1.02s 280ms 3.7x Incremental no-op (nothing changed) 995ms 278ms 3.6x
February 13, 2026 at 2:08 AM
davidcrespo.bsky.social
February 12, 2026 at 8:03 PM
davidcrespo.bsky.social
this is wonderfully put for the new yorker audience and exactly right. and, importantly, was always true even when the models weren’t that good at stuff
And sometimes they are. For a vanilla utterance like “The cat sat on the ___,” “mat” is a statistically better bet than “cummerbund.” If you can predict only the next word, however, it would seem impossible to say anything meaningful. When a model appears to do so, it must be cheating—by, say, repeating “embedded word chains.” But this view—that the models are only copying and pasting stuff they once read—cannot survive even a cursory interaction with them. On the tenth floor, Batson typed the prompt “A rhyming couplet: He saw a carrot and had to grab it,” and Claude immediately produced “His hunger was like a starving rabbit.” If the model were merely winging it one word at a time, like a “Looney Tunes” character who bridges a chasm by tossing out planks as needed, to land on a rhyme would be incredible luck. It’s not. When the model predicts the next word, it is not doing so just on the basis of the words that came before. It is also “keeping in mind” all the words that might plausibly come after. It predicts the immediate future in the light of its predictions of the more distant future.
February 12, 2026 at 1:48 PM
davidcrespo.bsky.social
lol joe gets it
February 12, 2026 at 2:21 AM
davidcrespo.bsky.social
wild angle to the downward revisions: it might mean productivity was higher than we thought
www.nytimes.com/live/2026/02...
Talmon Joseph Smith Feb. 11, 2026, 9:28 a.m. ETFeb. 11, 2026 Talmon Joseph Smith Matt Bush, an economist at Guggenheim Investments, noted that, “Today’s downward revisions to last year’s job numbers imply that productivity in 2025 was maybe even stronger than we realized and may be revised up.” It can sound somewhat counterintuitive, because the reduced job gains are a disappointment, but productivity is a function of business output per hour. And, for now, it appears U.S. businesses have been able to produce even more output, while adding far fewer workers than usual.
February 11, 2026 at 7:20 PM
davidcrespo.bsky.social
blocked again for truth. clearly I should not bother
February 11, 2026 at 7:06 PM
davidcrespo.bsky.social
xcancel.com/MattBruenig/...
February 11, 2026 at 1:58 PM
davidcrespo.bsky.social
his original definition is very funny
truthsocial.com/@realDonaldT...
Avatar Donald J. Trump @realDonaldTrump The United States has a chance to do something that should have been done DECADES AGO. Don’t be Weak! Don’t be Stupid! Don’t be a PANICAN (A new party based on Weak and Stupid people!). Be Strong, Courageous, and Patient, and GREATNESS will be the result!
February 11, 2026 at 5:30 AM
davidcrespo.bsky.social
my fully client-side BYOK static site LLM client (use at your own risk) now has a usable layout on desktop
github.com/david-crespo...
llm-web-spa.netlify.app
February 11, 2026 at 5:00 AM
davidcrespo.bsky.social
crushing it today
apples messages screenshot them: It’s a 2003 Brittany outfit me: it’s britney bitch
February 10, 2026 at 10:50 PM
davidcrespo.bsky.social
love to see it @boshen.github.io
github.com/oxidecompute...
Single file Whole repo (475 files) oxfmt (npx) 287.5ms ± 3.9ms (10 runs) 446.1ms ± 11.2ms (10 runs) prettier (npx) 405.2ms ± 3.4ms (10 runs) 3.213s ± 0.037s (10 runs) oxfmt 1.4x faster oxfmt 7.2x faster oxfmt (direct) 157.6ms ± 1.9ms (18 runs) 332.4ms ± 48.9ms (10 runs) prettier (direct) 283.0ms ± 1.5ms (10 runs) 3.107s ± 0.061s (10 runs) oxfmt 1.8x faster oxfmt 9.4x faster
February 10, 2026 at 10:01 PM
davidcrespo.bsky.social
how could I ever trust this process considering this extremely obvious failure mode which is easy to mitigate and which you've explained over and over that you did in fact mitigate
xcancel.com/MattBruenig/...
Matt Bruenig @MattBruenig 6h There are 1,233 citations that were written by the LLM. Every single one of them is hyperlinked to the document in the NLRB Research database. By forcing it to hyperlink all its citations, I could write a script to follow each link and make sure it is the same case being cited. I can define anything @2dionysus 6h Genuinely why would I gamble on if the citations are hallucinated or not in a “book” the “author” probably never even read fully to check before publishing? What’s the value of this? mattbruenig.com/2026/02/10/t… Feb 10, 2026 · 2:16 PM UTC Matt Bruenig @MattBruenig 6h There were a few mistaken citations, which were able to be detected through this process and fixed. So I can be sure there are zero hallucinated citations. That is simply not a possible failure point. Though it is important to actually build around these possibilities.
February 10, 2026 at 8:49 PM
davidcrespo.bsky.social
"The most significant limitations to this method come from contradictions and irrelevancies that exist inside the NLRA case law." 😁
nlrbresearch.com/book/sample
2.5 Limitations People who have some familiarity with Large Language Models often think that they are very prone to error in the form of “hallucinations,” i.e. making things up. There have been many reports of lawyers using LLMs to write briefs only to find, after they submitted the briefs, that the LLMs had cited non-existent cases and invented fake points of law. But these kinds of errors typically result from individuals typing legal questions into general-purpose LLM chatbots. This is completely different from how I am using LLMs for this book. Whenever I prompt Gemini or Claude to summarize specific cases, I also provide each LLM all of the text and information that it needs to do so. The prompt instructs the LLM to base its summary only on the text and information I have provided and to include citations with links back to the NLRB Research Database for each document it relies upon in the summary. Carefully prompting the LLM with all of the necessary context dramatically reduces the likelihood of a hallucination while forcing it to cite its sources with links makes it easy to double-check the output. The most significant limitations to this method come from contradictions and irrelevancies that exist inside the NLRA case law. The 100 most recent cases citing to a target case may not all be consistent with one another. In some instances, the older cases may have even been overruled by more recent decisions. Sometimes decisions cite to target cases for irrelevant or even incorrect reasons. These sorts of complications are difficult for an LLM to successfully disentangle. I have used my own expert knowledge to attempt to spot these kinds of errors and correct them when they appear
February 10, 2026 at 8:43 PM
davidcrespo.bsky.social
1956! killer rec from the robot
2 THE HUMAN USE OF HUMAN BEINGS The purpose of this book is both to explain the potentiali- ties of the machine in fields which up to now have been taken to be purely human, and to warn against the dangers of a purely selfish exploitation of these possibilities in a world in which to hum:m beings, human things are all- important. That we shall have to change many details of our mode of life in the face of the new machines is certain; but these machines are secondary in all matters of value that concern us to the proper evaluation of human beings for their own sake and to their employment as human beings, and not as second-rate surrogates for possible machines of the future. The message of this book as well as its title is the human use of human beings.
February 10, 2026 at 5:08 AM
davidcrespo.bsky.social
norbert wiener emphasizing that while it might sound like he’s saying these things are good, he is just trying to get his analysis right
WHAT IS CYBERNETICS? 1 HAVE BEEN occUPIED for many years with problems of com- munication engineering. These have led to the design and investigation of various sorts of communication machines, some of which have shown an uncanny ability to simulate human behavior, and thereby to throw light on the possible nature of human behavior. They have even shown the exist- ence of a tremendous possibility of replacing human be- havior, in many cases in which the human being is relatively slow and ineffective. We are thus in an immediate need of discussing the powers of these machines as they impinge on the human being, and the consequences of this new and fundamental revolution in technique. To those of us who are engaged in constructive research and in invention, there is a serious moral risk of aggrandizing what we have accomplished. To the public, there is an equally serious moral risk of supposing that in stating new potentials of fact, we scientists and engineers are thereby justifying and even urging their exploitation at any costs. It will therefore be taken for granted by many that the at- titude of an investigator who is aware of the great new pos- sibilities of the machine age, when employed for the purpose of communication and control, will be to urge the prompt exploitation of this new "know-how" for the sake of the machine and for the minimization of the human element in life. This is most emphatically not the purpose of the present book.
February 10, 2026 at 5:02 AM
davidcrespo.bsky.social
I think we are legit in an After Virtue/Canticle for Leibowitz situation with respect to accountability. we do not know what it means. its social meaning is dead
1 A Disquieting Suggestion Imagine that the natural sciences were to suffer the effects of a catastrophe. A series of environmental disasters are blamed by the general public on the scientists. Widespread riots occur, laboratories are burnt down, physicists are lynched, books and instruments are destroyed. Finally a Know- Nothing political movement takes power and successfully abolishes science teaching in schools and universities, imprisoning and executing the remain- ing scientists. Later still there is a reaction against this destructive move- ment and enlightened people seek to revive science, although they have largely forgotten what it was. But all that they possess are fragments: a knowledge of experiments detached from any knowledge of the theoretical context which gave them significance; pans of theories unrelated either to the other bits and pieces of theory which they possess or to experiment; instruments whose use has been forgotten; half-chapters from books, single pages from anicles, not always fully legible because torn and charred. Nonetheless all these fragments are reembodied in a set of practices which go under the revived names of physics, chemistry and biology. Adults argue with each other about the respective merits of relativity theory, evolutionary theory and phlogiston theory, although they possess only a very panial knowledge of each. Children learn by hean the surviving por- tions of the periodic table and recite as incantations some of the theorems of Euclid. Nobody, or almost nobody, realizes that what they are doing is not natural science in any proper sense at all. For everything that they do and say conforms to cenain canons of consistency and coherence and those contexts which would be needed to make sense of what they are do- ing have been lost, perhaps irretrievably. In such a culture men would use expressions such as 'neutrino', 'mass', 'specific gravity', 'atomic weight' in systematic and often interrelated ways which would resem… 2 After Virtue abound. Subjectivist theories of science would appear and would be criticized by those who held that the notion of truth embodied in what they took to be science was incompatible with subjectivism. This imaginary possible world is very like one that some science fiction writers have constructed. We may describe it as a world in which the language of natural science, or pans of it at least, continues to be used but is in a grave state of disorder. We may notice that if in this imaginary world analytical philosophy were to flourish, it would never reveal the fact of this disorder. For the techniques of analytical philosophy are essentially descriptive and descriptive of the language of the present at that. The analytical philosopher would be able to elucidate the conceptual structures of what was taken to be scientific thinking and discourse in the imaginary world in precisely the way that he elucidates the conceptual structures of natural science as it is. Nor again would phenomenology or existentialism be able to discern anything wrong. All the structures of intentionality would be what they are now. The task of supplying an epistemological basis for these false simulacra of natural science would not differ in phenomenological terms from the task as it is presently envisaged. A Husserl or a Merleau-Ponty would be as deceived as a Strawson or a Quine. What is the point of constructing this imaginary world inhabited by fic- titious pseudo-scientists and real, genuine philosophy? The hypothesis which I wish to advance is that in the actual world which we inhabit the language of morality is in the same state of grave disorder as the language of natural science in the imaginary world which I described. What we possess, if this view is true, are the fragments of a conceptual scheme, pans which now lack those contexts from which their significance derived. We possess indeed simulacra of morality, we continue to use many of the key expressions. But we have-ve…
February 9, 2026 at 9:04 PM
davidcrespo.bsky.social
an engineer can never be held accountable, therefore an engineer must never make an engineering decision
February 9, 2026 at 8:03 PM
davidcrespo.bsky.social
meanwhile on twitter
paperclippriors @paperclippriors · 2m So jarring watching the Super Bowl and hearing things like “this coverage is brought to you by Claude, keep thinking”. Does the average person even know what that means? Link @LinkoftimeG · 2m Superbowl ads be poppin off lmao #claude that cougar ad was freaking hilarious next to the #instacart #SuperBowl
February 9, 2026 at 12:02 AM
davidcrespo.bsky.social
enjoying the real time reactions to the claude ad
‪David Bailey‬ ‪@drgdave.bsky.social‬ · 4m Watching the Super Bowl just now, I learned that the tagline for Claude AI is "Keep Thinking", which is a ridiculous non-sequiteur. It's like DoorDash using the tagline: "Keep cooking from scratch". ‪Wendy Sparrow‬ ‪@wendysparrow.bsky.social‬ · 5m I can't believe the Super Bowl sponsor ad slogan for Claude is "Keep thinking." Given the vast amount of evidence coming back showing that Gen AI users stop learning and lose skills... ‪Geoffrey Card‬ ‪@rangutang.bsky.social‬ · 5m Replied to alyssa mercante I love that the Claude ad, in trying to take down ChatGPT, successfully made ALL AI look like a nightmare, including Claude. ‪beef boy‬ ‪@beefboy.bsky.social‬ · 5m claude announced as official thinking partner of NFL after numerous brain injury scandals ‪hailey‬ ‪@hailey.at‬ · 9m super bowl sponsored by claude is so funny lol
February 8, 2026 at 11:59 PM