Post

I covered this in my premium last week, it’s laughable! www.wheresyoured.at/ai-mythbuste...

Kat Tenbarge@kattenbarge.bsky.social

I find the idea that you need to know “how to use” AI tools really funny, as if it isn’t incredibly easy to type in prompts and the marketing these tools is that literally anyone can do it

177 reposts978 likes

“If you’re not using AI for your day-to-day tasks, then you need to do more training and are simply not dedicated enough! Remember - the issue is definitely you”.

"AI will be really good really soon, you need to learn how to use it or you'll be left behind" "cool, how will i be using it then?" "oh, typing a plain language sentence" "i can do that when it's ready, i'm good" "no no you need to invent demand for it now so they have the money to do that"

like if excel was first developed as not being able to do math and they said "buy it anyways so MS has the funding to make it do math right eventually" theyd be laughed out of the fucking room

So this "spreadsheet" thing. What should I do with it? We have no idea but we just spent a billion dollars and had to lay off 20% of the company to afford it so you need to figure that out for yourself Totally imaginary conversation if Excel was sold like Copilot.

I like to believe i spend a good amount of time learning a lot of skills for fun and work, every prompt engineering course seems to be extremely simple to learn. its trial and error + asking for tips. I think its better to learn skills without ai, and then use it, then spend time "learning" to us ai

Also a journalist here. I've tried out every LLM since GPT 2 for viability as a research-assistant. They suck. Reasons: They have no idea of relevancy. They omit interesting anecdotes, they will miss patterns of behaviour etc. They will choose the wrong focus. On top of that come the hallucinations.

You can only use them if mediocre is good enough for you. And even then you will have to still fact-check everything. You'll also be left with a lessened understanding of the story you're trying to report. It sounds good and the latest versions initially seem like they could be valuable. But: nope.

The defining feature of LLMs is that they’re a structured “dark pattern” meant to launder the idea that a mind’s on the other side. To that end, unlike any good technology, there’s no affordance that promotes any intended use. Physical objects like cups and hammers have handles, for example.

Even open-ended digital interfaces have input validation + confirmation of errors. But LLMs have anti-UI. Users are encouraged to ask everything, including what LLMs can’t do, then must validate. Every part of this UX offloads work on to the user, engineers, annotators to prop up the mind illusion.

As a software engineering manager, I’ve observed that some of my directs use AI more effectively than others. It’s not that they write better prompts; rather, they have better discipline around testing and reviewing the code generated by the AI.

Peers and I just avoid. Between switching tech stacks constantly, tons of contextual stuff per layer no one has time to explain to AI, decades in languages and patterns, plus coding on/past the edge of known performance and behavior, we've no time for it. PowerBI kids are all in tho.

I think that’s a fine choice. The point I’m trying to make is that it is, in fact, possible to use AI well and also possible to use it poorly. So when I hear folks laugh about “learn how to use AI” as though there’s nothing to learn, I feel like there are, actually, some things to learn.

I cringe when by skills, what we mean is edit someone else's code (AI this time) since most devs are worse editors than they are creators. Heavily leaning on test scaffolding will help, assuming they are good at it. Too often it's gonna streamline single positive result devs to pumping junk.

I use Resharper AI Assistant with Visual Studio and it really does increase my productivity substantially. You have to double check everything it does, though, because it has a propensity for just making stuff up.

Training org I work for has been looking at possibly delivering a Diploma in Artificial Intelligence, that's now a registered accredited course. Not just a single unit of competency on using the tools for business, but a whole diploma on top of your business qualification

I have discovered exactly one use case for LLMs in my job: feed in a long transcript of a recorded call, and use the LLM as essentially a search engine that can search by topic, rather than keyword. It turns 16 hour tasks into…15.5 hour tasks.

Maybe AI is the perfect bullshit technology for bullshit jobs (as described by David Graeber)? i.e. I've heard some academic say it's useful for doing formulaic paperwork that is performative rather than purposeful i.e. University management demands for bullshit filed and forgotten audit trails.

AI isnt, it is nothing more than a programmed response to query, with better databases than previous ones. I have seen better "ai" in video games where it learns your patterns and accounts for them