iiinova — where your digital self keeps evolving

Reposted by Butlerian Jihadist of House Dornick

Kip Williams sort of @kipwilliams.bsky.social · 22m

As we perfect perpetual motion, we will no longer need a kicker battery on the side to keep it working.

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 12m

People get wealth fucked up. You only need to be lucky or have a really good idea *once* to be wealthy the rest of your life. You dont need to have another good idea the rest of your life to stay wealthy. Stop thinking these mfs are smarter than you.

0100

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 19m

Most of the belief in the potential of LLMs comes from blind faith in rich people and an inability to process that rich people are capable of being incompetent.

141

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 41m

Notice how all the benefits of LLMs are always framed in terms of hypotheticals and appeals to a future version of the tech that may not come to exist in respose to empirical evidence of how they are *actually* used *today.

040

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 43m

The study methodology covers +700 companies on Jellyfish a codebase management platform tracking the work of 700,000 programmers based on their Bitbucket and Github activity as reported to Jellyfish. We have a proper sample size of people actually using these tools.

050

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 47m

AI winter has happened more than once before, we've had several tech bubbles and there is an entire graveyard of technology we thought had promise but flopped to the loss of billions of dollars. Welcome to Earth. In your orientation packet on page 7 you will find the pie selections.

140

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 50m

We know its improved significantly....at producing shit code faster which is the point and due to the design *concept* of the tool, we have reason (and evidence) to believe code quality is being traded for code productivity and it will worsen over time.

000

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 58m

They arent language models. Those are harnesses of classical algorithms bolted onto LLMs and the study accounts for those. We still see no improvement. The study uses real world data from Jellyfish. It captures users who use Claude Cowork too.

110

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 1h

That assumes the tool *can* be used for that purpose with the constraints of its design concept and we already know hallucinations are *inherent* to LLM design. That means there can be no way to solve this problem by design. The solution will require a different technology.

100

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 1h

In engineering all design occur *within* a paradigm that both enables and constrains the design concept. That means *all* design concepts have a ceiling for improvement. The design concept for a wheeled passenger vehicle will never "improve" to become a Millienium Falcon.

051

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 1h

An elephant will never get better at climbing trees in an timescale that matters.

020

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 1h

Its had enough time. The notion that will improve is based solely on faith. Historically technology improves at a given task over time is when the flaw lies implementing or lack of optimization. It *never* improves when the flaw is inherent to the design concept.

250

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 1h

There is no evidence of that. Its not a matter of improvement because the problem is due to the stochastic nature of the system. The flaw is *inherent* to LLMs as a concept. This is like saying a butter knife will eventually become a lightsaber if I keep sharpening it.

0122

Reposted by Butlerian Jihadist of House Dornick

Devon Greyson @greyson.bsky.social · 9h

Sorry, did someone say grant applications and scientific manuscripts? #SameProblem #MoreFasterIsNotBetter

2359

Reposted by Butlerian Jihadist of House Dornick

Jak @jaksays.bsky.social · 7h

LLMs make the typing bit easy But the typing bit is the simple part of development - the hard parts are requirements gathering, problem analysis and verification, and as per this study those are parts LLMs actually make harder Those are also the only parts that add business value

2323

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 7h

This is the same issue with the hiring process and TA too now that I'm thinking about it: Higher volumes of shittier resumes that TA has to wade through which makes the entire process bad for everyone involved.

241227

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 8h

When are we gonna stop talking about what we could, should or would do with AI and start actually doing it? Its been years and several trillion dollars spent. The time to speculate on its potential is over. Show me the money.

1100

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 8h

The pace of model performance as a function of compute began to plateau almost a year ago...the models arent actually improving much at all...which is why you still see no evidence *anywhere* they are delivering on any of their promises to users.

150

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 8h

This is why when its sold to you they are always talking about vague notions of "efficiency" and "productivity" but they *never* discuss outcomes. Sure its more efficient and productive....at producing dogshit lol.

090

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 8h

The tools dont produce *optimal* code as easily as a senior engineers doing it by themselves. They tend to produce way more lines of code than is necessary for a function.

2170

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 8h

Depends on the size of the code base and complexity of the solution but so far that doesnt seem to make a difference. The bottleneck is still reviews and revisions even with senior level engineers because of the volume of code produced. And senior devs have to spend time more time refactoring.

1150

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 8h

Its an average. Some were actually worse lol.

1110

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 8h

Its FOMO and game theory fueled by pure ignorance of the technology by management. Its the same reason why all these companies are burning billions on something that *clearly* wont work like they thought.

0491

Reposted by Butlerian Jihadist of House Dornick

George W Harris @bullcityfats.northsky.social · 9h

Junior engineers also eventually become senior engineers, which Claude never will.

2252

Reposted by Butlerian Jihadist of House Dornick

Anthony Moser @anthonymoser.com · 9h

it's good to see this laid out clearly, but i'm once again going to point to Peter Naur's 1985 essay "Programming as Theory Building"

The present discussion is a contribution to the understanding of what programming is. It suggests that programming properly should be regarded as an activity by which the programmers form or achieve a certain kind of insight, a theory, of the matters at hand. This suggestion is in contrast to what appears to be a more common notion, that programming should be regarded as a production of a program and certain other texts.

621361

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 10h

Gen AI isnt making any money and wont because the *one* field it has the greatest application, software engineering, none of the businesses deploying it are seeing any productivity benefit that translates to P&L.

0101

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 10h

Facts. I'm paid the big bucks for the edge cases and solving problems that dont have an existing solution yet.

010

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 10h

Academic research papers generated with AI are actually *worse* than software generated with AI because shitty code can give you immediate feedback in the form of bugs and code that wont compile. Shitty academic papers can take time to sniff out.

2172

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 10h

Yep which I believe will cause software engineering as a *field* to deteriorate over time.

61493

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 10h

The software engineers that now report to me told me a while ago that AI tools can get you 90%-95% there but that remaining 5%-10% kills all the productivity you gained.

4110

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 10h

Every single study done on Gen AI and worker productivity over the past 3 years, from multiple countries, has come to exact same conclusions and none of the improvements in the models over that period of time has put a dent in the problem.

1102

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 10h

It wouldn't have made a difference given what they measured. They were measuring *all* the enterprise coding tools being used using Jellyfish data not just Claude.

110

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 10h

The other thing is tech worker payroll still higher than it was before covid 🙃. Which means the AI job apocalypse is bullshit like I been saying for a year.

050

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 10h

The best part about having experience in a field is you can predict the mistakes less experienced people in the field will make before they make them.

020

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 10h

The average increase on code volume after the adoption of agentic AI studying 700 firms is 4,000 additional lines of code per programmer. The advantage of junior engineers writing code without the use of AI is senior engineers had to spend less time on code reviews.

876779

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 10h

Yep and this is why virtually no one is seeing any actual value out of it and none of the AI companies are making any money.

1470

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 10h

Basically AI allows software engineers to produce shitty code *faster* and in higher volumes which increases the amount time spent on code reviews and revisions nullifying the productivity gained from producing the code in the first place.

171502263