Sign in

conputer dipshit

@davidcrespo.bsky.social
5.3K followers 652 following 19K posts

web dev + hot dad. enjoy charts, unions, conputer games, philosophy. chicago crespo.business

PostsRepliesMedia
conputer dipshit @davidcrespo.bsky.social · 3m
point for my strategy of never reading the chain of thought. you might as well ask the model why it did what it did, the answer‘s not going to be any less bullshit
051
Reposted by conputer dipshit
Kiran @kirancodes.me · 2h
category theorists don't want you to know that category theory is dead simple if you are allowed to visualise it in 3d tear down the category cabal now!!!
0446
Reposted by conputer dipshit
Al Merose @al.merose.com · 46m
If the workers of Ant, OAI, and GDM were all part of a union (e.g. a frontier AI guild), they could cooperate in ways that are illegal collusion for corporations. IMO, that would be a stronger alternative but achieve similar goals (slowdown for safety) of a cartel.
1294
conputer dipshit @davidcrespo.bsky.social · 2h
in anticipation of mitchellh's superlogical terminal multiplexer (demos: x.com/mitchellh) I have vibed a CLI for manipulating ghostty through applescript this lets agents split a pane to show a live task tracker in a markdown file with glow (github.com/charmbracele...) github.com/david-crespo...
Terminal screenshot showing the help output and layout inspection for a CLI tool named gpane:

$ gpane -h

Usage: gpane

Description:
  Inspect and split the Ghostty panes of the tab this runs in.

Options:
  -h, --help  - Show this help.

Commands:
  ls                                 - Show this tab's panes and their inferred layout.
  split   <pane> <direction>         - Split a pane in this tab, optionally running a command in the new pane.
  resize  <pane> <edge> <position>   - Move one edge of a pane to N% of the tab, measured from the left or top.

$ gpane ls
window 1, tab 1 "dotfiles | Update the Glow checklist tracker format": 4 pane(s)
layout: (1 / 2) | (3 / 4)

#  id        x,y%   w x h%  cols x rows  cmd     cwd
1  55CA6D71  0,0    43x35   75x18        glow    ~/repos/dotfiles
2  231942CD  0,35   43x65   75x35        claude  ~/repos/dotfiles
3  90CC96E6  43,0   57x50   99x27        hx      ~/repos/dotfiles
4  E25F85BE  43,50  57x50   99x27        deno    ~/repos/dotfiles  self,focusedTerminal window showing a markdown file rendered in Glow with a bottom status bar and command output.

Text content:

Tab title: dotfiles | Add a note to the Glow checklist tracker ⌘1
Background tab: console | Assess lzx and xy branch split while fixing...

gpane simplification

Goal: trim bin/gpane.ts (rev qzksvqzu) without changing behavior. Work in revs on top for review. 3, 4, 5, 7 are squashed into qzksvqzu.

## Candidates

[✓] 1. Resize new split with resizeSplit (tylzyppm)
• --size path calls resizeSplit after the split instead of the split AppleScript doing it. -14 lines. Tested a 30% split live.
[✓] 3. axis() helper (squashed)
• Replaces the repeated up || down / left || right checks. Also dropped an unused extent in planResize.
[✓] 4. Shared combine() for size arithmetic (squashed)
• Used by candidates() in inferLayout and by treeSize.

Glow status bar: Glow .claude/notes/2026-09-30-gpane-simplify-tracker.md 0% ? Help

Command output below:
Wrote 56 lines to .claude/notes/2026-09-30-glow-tracker-list-rendering.md
351
conputer dipshit @davidcrespo.bsky.social · 4h
the critique that the labs can't get training costs under control doesn't really land because they are not aiming to maximize profits. they are aiming to maximize the long-term growth of training compute, which incidentally requires just enough profits to keep money flowing
271
conputer dipshit @davidcrespo.bsky.social · 4h
fun — write macos shortcuts in code rather than having to use the GUI like a psycho github.com/electrikmilk...
Code editor showing a Cherri script on the left and compiled Apple Shortcuts blocks on the right under a yellow smiley face icon titled "Hello Cherri".

Cherri code:
// Hello, Cherri!

#define name Hello Cherri
#define glyph smileyFace
#define color yellow

@message = "Hello, Cherri!"
alert("{message}", "Alert")

Compiler output:
Text: Hello, Cherri!
Set variable message to Text
Show alert message, Title: Alert, Show Cancel Button: off
1101
conputer dipshit @davidcrespo.bsky.social · 5h
we'll see if they can break the trend of benchmarking well, horrible vibes, and then being surpassed on the benchmarks too within a month
7564
Reposted by conputer dipshit
Eric Columbus @ericcolumbus.bsky.social · 12h
Not a great sign when your 12-person private fundraiser includes someone who hates you enough to leak damaging audio
The audio of Mr. Paxton was shared with The Times by an attendee who recorded him speaking at a private fund-raiser in Washington attended by roughly a dozen people. The person, who opposes Mr. Paxton’s campaign and provided the recording to The Times, was granted anonymity because Mr. Paxton’s remarks were private.
562489497
Reposted by conputer dipshit
Aster ID @aster.id · 12h
The modern research ecosystem of online tools and communities is controlled by for-profit companies. But we think that researchers should be the priority - not the product. So, we're announcing Aster ID: not-for-profit Atmosphere account hosting for researchers.
blog.aster.place
Announcing Aster - a home for researchers on the open web
13546118
Reposted by conputer dipshit
conputer dipshit @davidcrespo.bsky.social · 10h
I think this is still quite relevant www.dreamsongs.com/Files/Patter...
Habitability and Piecemeal Growth
The sun paints the scattered clouds red, paints the hills across the Bay
a deeper red. Moment by moment the sun’s light flares windows in
the hills, moving upward as twilight approaches. The programmer
looks up from his screen to watch. He sees headlights moving one way
and taillights the other along either side of the highway spine. Home-
lights click on, and the Bay is a silhouette. His daughter: Will she be
too petrified to dance? Will she miss the cue? If only he could help her
pass by her fear. Reaching to the right of the mouse, he lifts his Coke
and drains it, saves his buff ers, drifts his eyes to the sensual move-
ment of the highway, pausing his mind as it passes over a detail in the
code he’s just typed. The thought passes before crystallizing, so he puts
the CDs in his blue Kelty pack—and heads for the recital.
Code is written by people. And people have a lot on their minds, they have lives to
lead, and those lives intrude on their thinking. You might pay a developer for a
full eight-hour day, but how much of that day is truly yours? Of course, not all of
us are software development managers, but the concerns of these managers
should rightfully be concerns of ours—concerns of language designers. The
astute software manager knows that just about every programmer except the
pathological workaholic trades his hours of overtime for hours of undertime. The
manager goes all out to ensure that the needs of the developer are met—home
terminals, sunny offices, minimal meetings, meals during the big push, and a
healthy ego boost from time to time.
If the software manager knows that people and their habits are the determin-
ing factor in software development, why don’t language and system designers?
Here is what Stroustrup says about the design of C++:
C++ is a general purpose programming language designed to make
programming more enjoyable for the serious programmer. (Strous-
trup 1987)
2133
conputer dipshit @davidcrespo.bsky.social · 10h
I’m going with the critique of pure reason
1563
conputer dipshit @davidcrespo.bsky.social · 10h
this is now an autocatalytic networks stan account
130
conputer dipshit @davidcrespo.bsky.social · 11h
very interesting charts in here www.apricitas.io/p/ai-and-the...
Stacked area chart titled "Video, % of US Consumer Spending" with the subtitle "Video Sales Fell After 2016, With Streaming Unable to Compensate for Cable & Movie Losses," tracking data from 1960 to the mid-2020s. 

Total spending on video grew from around 0.3% of consumer spending in 1960 to a peak near 1.1% in the early 2010s, before falling back toward 0.7% by 2024. 

The breakdown shows:
- Movie theater tickets declined from about 0.3% in 1960 to around 0.1% or lower from the 1980s onward, with a sharp drop in 2020.
- Cable, satellite, and other live TV expanded dramatically from 1980 through the 2000s, reaching roughly 0.8% of spending, before experiencing a steep decline post-2016 to around 0.4%.
- Video discs, tapes, and permanent downloads grew from the 1980s to the early 2000s, peaking around 0.25%, then largely vanishing by the 2020s.
- Video streaming and rental emerged around 2010 and grew to roughly 0.3% by the 2020s, but not enough to offset the drops in cable and physical media.
1200
conputer dipshit @davidcrespo.bsky.social · 12h
breaking news: there are stupid assholes on twitter this guy’s job is literally “AI grifter”. his opinion is of no consequence whatsoever. “oh but it represents an attitude“ ok, find me someone who isn’t a professional grifter saying it nicksaraev.com
9735
conputer dipshit @davidcrespo.bsky.social · 12h
nteresting read focused on refuting a folk theory of mind that ties thinking too closely to conscious experience. I think I agree with it overall, though I don’t think the folk theory is the only reason you might want to deny that LLMs are “really” thinking
3160
Reposted by conputer dipshit
Joey Politano🏳️‍🌈 @josephpolitano.bsky.social · 29/09/2026
Solar power made up 11.3% of US electricity this July, up from 9.5% last year Over the last twelve months, solar has made up 9.5% of US electricity, another record high!
a graph of solar's share of US electricity
221127
Reposted by conputer dipshit
MONOMOON | Gentle Giant Games @gentlegiantgames.bsky.social · 15/09/2026
MONOMOON, my non-linear game about exploration and discovery launches on Steam November 11th! With no proper instructions, you must deduce and achieve your goals as you solve puzzles, find mysterious items, and uncover secrets hiding right under your nose.
13426134
Reposted by conputer dipshit
Ned Resnikoff @resnikoff.bsky.social · 29/09/2026
I think about this interview with the distinguished philosopher Conan O'Brien a lot. I actually think this is a healthy way to live your life! www.nytimes.com/2019/01/14/a...
Is this how you want to go out, with a show that gets smaller and smaller until it’s gone?

Maybe that’s O.K. I think you have more of a problem with that than I do. [Laughs.] At this point in my career, I could go out with a grand, 21-gun salute, and climb into a rocket and the entire Supreme Court walks out and they jointly press a button, I’m shot up into the air and there’s an explosion and it’s orange and it spells, “Good night and God love.” In this culture? Two years later, it’s going to be, who’s Conan? This is going to sound grim, but eventually, all our graves go unattended.

You’re right, that does sound grim.

Sorry. Calvin Coolidge was a pretty popular president. I’ve been to his grave in Vermont. It has the presidential seal on it. Nobody was there. And by the way, I’m the only late-night host that has been to Calvin Coolidge’s grave. I think that’s what separates me from the other hosts.

I had a great conversation with Albert Brooks once. When I met him for the first time, I was kind of stammering. I said, you make movies, they live on forever. I just do these late-night shows, they get lost, they’re never seen again and who cares? And he looked at me and he said, [Albert Brooks voice] “What are you talking about? None of it matters.” None of it matters? “No, that’s the secret. In 1940, people said Clark Gable is the face of the 20th Century. Who [expletive] thinks about Clark Gable? It doesn’t matter. You’ll be forgotten. I’ll be forgotten. We’ll all be forgotten.” It’s so funny because you’d think that would depress me. I was walking on air after that.
132392513
conputer dipshit @davidcrespo.bsky.social · 23h
guess we’ll find out soon enough. this year’s college seniors have had chatgpt their entire time in college. this year’s junior devs have had it their entire careers
31018
conputer dipshit @davidcrespo.bsky.social · 30/09/2026
8yo asked a 4th grader friend what they were reading and it was The Martian. is that allowed?!
360
Reposted by conputer dipshit
grace @gracemachine.bsky.social · 29/09/2026
:) i love rain world so much... insane that i've gotten to work on two book length projects.
68231
Reposted by conputer dipshit
Isabella M. Weber @isabellamweber.bsky.social · 29/09/2026
"To save democracy, democratize the economy, admonishes economist Weber in her outstanding debut." This is the first advance review @publisherswkly.bsky.social. Humbled it’s starred and moved by how concisely it captures the message of ANTI-FASCIST ECONOMICS ❤️
46818
Reposted by conputer dipshit
Dave Weigel @daveweigel.bsky.social · 29/09/2026
I'm not the audience for the Smith hearing, but the main political issue here is that Republicans want the world to know that Smith subpoenaed congressional phone records as he investigated Jan. 6 etc, and nobody outside the Senate can explain to you why that's bad.
28934129
conputer dipshit @davidcrespo.bsky.social · 29/09/2026
I should say, I think the Clawlike category is totally going to happen now and it's mostly a matter of the models getting good enough at the price they need to be. like almost everything is: I am sort of an avowed extremist on the question of which problems reduce to model capability
1190
conputer dipshit @davidcrespo.bsky.social · 29/09/2026
if you post charts I WILL follow you
0150
Reposted by conputer dipshit
Guy Berger @econberger.bsky.social · 29/09/2026
Indeed job postings: 1/ More good news about the labor market, as of the 2nd half of September (so possibly more relevant to the October jobs report, than to Friday's September jobs report) After a micro-slump over the summer, postings regaining their spring momentum
1103
Reposted by conputer dipshit
Guy Berger @econberger.bsky.social · 29/09/2026
2/ Software development job postings growing at their fastest pace in a long time (>21% Y/Y), albeit from a big hole. On track for return to pre-pandemic levels by mid-2028.
1143
conputer dipshit @davidcrespo.bsky.social · 29/09/2026
good read, agree with a lot of it (except the idea that meta can succeed at making something good)
1100
conputer dipshit @davidcrespo.bsky.social · 29/09/2026
you get a jev! you get a jev! (OpenAI just announced a "Decisions API" which is just Luna w/ structured output and some optimizations)
121302
conputer dipshit @davidcrespo.bsky.social · 29/09/2026
nice review of studies on labor market impact of AI. nothing too new, but some nice charts. "Apart from early career workers in exposed fields, the impact of AI on the labor market has thus far been muted. That being said, it would be unwise to assume that this trend will continue."
aleximas.substack.com
Has AI impacted the labor market yet?
Evaluating the state of the evidence
1334
Reposted by conputer dipshit
Jeremy Diamond ☕️ @dmnd.me · 29/09/2026
Conputer Dipshit Batsignal
0151
conputer dipshit @davidcrespo.bsky.social · 29/09/2026
my god. don't watch a short clip. watch the whole thing
190
Reposted by conputer dipshit
Jack Jenkins @jackjenkins.me · 29/09/2026
Genuinely engrossing reporting here from the Times. Some of this is a re-tread of stuff that has already been reported, but the extensive interviews with Olah and others inside Anthropic is all very new, and deeply revealing. www.nytimes.com/2026/09/29/u...
nytimes.com
Is Claude Conscious? Inside Anthropic’s Quest to Instill Morality Into Its A.I. Models
In a series of private meetings, the company consulted religious scholars to help instill morality into its A.I. models — and make the case that Claude could be conscious.
5238
conputer dipshit @davidcrespo.bsky.social · 29/09/2026
this is really good
youtube.com
in the future
YouTube video by bill wurtz
050
conputer dipshit @davidcrespo.bsky.social · 29/09/2026
unfortunately it calls itself America, which is very confusing: "That is a policy question for Congress and the President, not something America can decide or argue."
Screenshot of a chatbot on America.gov:

why can't we have national healthcare system?

I don’t take a position on whether the United States should have a single national health system. That is a policy question for Congress and the President, not something America can decide or argue.

The federal government already runs several coverage programs. People typically get insurance through a mix of these:

Job-based plans — many people get coverage through an employer.
Health Insurance Marketplace — shop and enroll at HealthCare.gov. Open Enrollment for 2027 coverage
0101
Reposted by conputer dipshit
michael lutz @ztul.bsky.social · 29/09/2026
the Toy Story Peoples is a collective name given by archeologists to the powerful 21st century civilizations most notable for producing Toy Stories
39721
conputer dipshit @davidcrespo.bsky.social · 29/09/2026
the former gifted kids of bsky are going to love this lol. “we only make students write because we don’t have time to talk to them”
2400
conputer dipshit @davidcrespo.bsky.social · 29/09/2026
cool paper, cool tables
Table titled: "Table 1: Types of 'administrative burden' citizens face when interacting with government (Moynihan, Herd, and Harvey 2015), the AI agent capabilities that may reduce them, and technologies that enable these capabilities."

Columns: Burden Type | Specific Burden | Agent Capability | Enabling Technologies

Burden Type: Learning Costs
- Specific Burden: Understanding complex rules and procedures | Agent Capability: Summarizing complex regulations and requirements in plain language | Enabling Technologies: LLMs with large context
- Specific Burden: Uncertainty about eligibility | Agent Capability: Evaluating personal data against policy and entitlement rules | Enabling Technologies: Context augmentation, directly or, e.g., via Model Context Protocol (MCP)
- Specific Burden: Unawareness of services or entitlements | Agent Capability: Scanning or continuously monitoring for entitlements; proactive alerts | Enabling Technologies: Web search tools; pipelines with scheduled execution

Burden Type: Compliance Costs
- Specific Burden: Legal writing; filling forms | Agent Capability: Context-specific, sophisticated writing | Enabling Technologies: LLMs with user context
- Specific Burden: Confusing government websites and service portals | Agent Capability: Navigating and interacting with websites autonomously | Enabling Technologies: Browser use tools; vision-language models
- Specific Burden: Locating and assembling supporting documents | Agent Capability: Retrieving, formatting, and attaching required supporting materials | Enabling Technologies: File system access; multimodal document parsing
- Specific Burden: Managing response deadlines and follow-ups | Agent Capability: Monitoring deadlines; alerting users or submitting replies autonomously | Enabling Technologies: Asynchronous task queues; calendar API integration

Burden Type: Psychological Costs
- Specific Burden: Having to re-explain case context to many contacts | Agent Capability: Maintaining full ca…Table summarizing possible government responses to agentic flooding, grouped under two strategies.

Table contents:

Columns: Strategy, Approach, Sample Methods, Benefits, Drawbacks.

Row 1:
- Strategy: Suppress Demand
- Approach: Add Friction
- Sample Methods: Fees; In-person appointments; Access limits, e.g. rate caps; Closing digital channels
- Benefits: Fast to deploy; Few prerequisites; Proven to work
- Drawbacks: Can drive procedural inequality; Worsens user experience; Some forms may erode as AI capabilities advance

Row 2:
- Strategy: Suppress Demand
- Approach: Reduce the Expected Benefit
- Sample Methods: Reduce entitlement amounts; Narrow eligibility criteria; Fines for unsuccessful applications
- Benefits: Robust to advances in agent capabilities; Policy option usually established
- Drawbacks: Legislative decision, can trade off other priorities; Does not always address operational impact

Row 3:
- Strategy: Increase Capacity
- Approach: Redesign the Service
- Sample Methods: Structured interfaces with identity verification; Deterministic processing pipelines; Proactive "push" delivery
- Benefits: Technical architecture is established; Can also improve citizen experience
- Drawbacks: High upfront investment; Cross-agency coordination required; Often needs legislative change, can take years

Row 4:
- Strategy: Increase Capacity
- Approach: Deploy Agents in Processing
- Sample Methods: Intake and triage; User correspondence; Autonomous processing of routine cases
- Benefits: Scalable, likely effective; Improves with capabilities improvements
- Drawbacks: Legal limits on automated decision-making; Provider dependency and sovereignty risks; Ethical and societal risks

Caption: Table 3: Overview of possible government responses to agentic flooding, grouped under two strategies.
092
Reposted by conputer dipshit
Forest Gregg @bunkum.us · 13/06/2026
@rnrkd.bsky.social nailed it the vast majority of these cases are brought by employees, the increase is also cases in employees. they are overwhelmingly pro-se, and there is no increase in proportion of representation. charges against the teamsters and unite-here make up about 30% of the growth.
222
conputer dipshit @davidcrespo.bsky.social · 29/09/2026
nothing really new or interesting in here, sadly. all the numbers are for 2025 and are dwarfed by everything that has happened since
reuters.com
EXCLUSIVE: Anthropic's IPO prospectus shows sweeping AI vision, surging costs
Anthropic is making a massive bet that AI will ​transform the global economy more profoundly than industrialization, electricity and the internet, according to its IPO prospectus seen by Reuters. The ...
2580
conputer dipshit @davidcrespo.bsky.social · 29/09/2026
kinda weird that this article (and this other one www.nytimes.com/2026/09/23/u...) has nothing to say about LLMs. where do you think they got a zero-day in Oracle PeopleSoft?
nytimes.com
Embarrassing Breach at F.B.I. Fuels Fears of Harm to Its Employees
In an internal memo, the F.B.I. said it believed the hackers may have stolen sensitive information on all of its employees.
1180
Reposted by conputer dipshit
Luke O'Neil @lukeoneil47.bsky.social · 29/09/2026
Stunning results as reported by thew French newspaper Le Pédophile Raciste
21539
Reposted by conputer dipshit
Frank Lantz @flantz.bsky.social · 28/09/2026
CONIMG SOON
37617
conputer dipshit @davidcrespo.bsky.social · 28/09/2026
very true that conceiving your activity as a set of tasks agents can do is hard, even for highly sophisticated computer users, even programmers, even people who are actively using agents all day. outside of work my LLM use is almost exclusively as fancy web search
9683
Reposted by conputer dipshit
Matt Keeter @mattkeeter.com · 28/09/2026
hi friends I'm very excited to announce Halfspace, my experimental cross-platform IDE for solid modeling with distance fields! This has been in progress since midway through last year, and it's finally "done enough" to make it more public.
mattkeeter.com
Halfspace
An IDE for solid modelling with distance fields
520648
conputer dipshit @davidcrespo.bsky.social · 28/09/2026
what world is this? taken out of context I would have assumed this is a parody blog.alexewerlof.com/p/coding-is-...
We know for a fact that even the most sophisticated LLMs aren’t fully capable of “common sense”. Humans on the other hand:

    Understand the non-verbal communication and unstated intentions better than LLMs

    Naturally push back until a mutual understanding is achieved.

    When wrong, they’re consistently wrong, meaning they don’t have “jagged intelligence”

    When right, they are [typically] right and continue to operate at an expected level (until fatigue hits but that’s different from AI flip flopping between success/failure).
3250
Reposted by conputer dipshit
missingjigsaw.bsky.social @missingjigsaw.bsky.social · 28/09/2026
1. Glonzo is everything that is the case. 1.1 Glonzo is the totality of things, not of facts. 1.2 Glonzo is not determined by the facts, nor by these being *all* the facts. 1.3 For the totality of facts cannot determine what is Glonzo, nor also all that is not Glonzo.
15310
Reposted by conputer dipshit
Matt Hodges @matthodges.bsky.social · 28/09/2026
The agents could apparently only send GET requests, but the UN endpoint they wanted only accepted POSTs. They routed POSTs through a third-party scanning service, and they also used a URL-encoding trick that got the server to answer GETs it was supposed to reject. swarmcha.se/posts/openai...
swarmcha.se
OpenAI agents tried to bruteforce a UN website's API fields
From 13 April - 19 June 2026, OpenAI agents scanned UNCTAD's API ~16,500 times, using proxies, obfuscation, and Google's XSS game · By Rowan H-J (LinkedIn)
092
Reposted by conputer dipshit
Daniel Denvir @danieldenvir.bsky.social · 28/09/2026
Ben and Quinn do an unbelievably good job guiding us through our bizarre political-economic moment. Listen.
0112
Reposted by conputer dipshit
Don Moynihan @donmoyn.bsky.social · 28/09/2026
New, from me: Once memory-holed, DOGE is now the subject of revisionism. To better understand why DOGE failed, I talked to people in the civic tech movement that DOGE displaced, who continue to offer the alternative model for how to modernize govt. 🧵 donmoynihan.substack.com/p/doge-vs-ci...
donmoynihan.substack.com
DOGE vs. Civic Tech
As DOGE revisionism picks up, a contrast with the civic tech movement it displaced illuminates its failure
18792258