Give me your hottest takes:

[scott@sfba.social]

@annaecook sigh

My feeling (not sure it even counts as a “take”) is that the fantasy and dream of what “AI” *could* be is too enticing to resist.

There is admittedly an “indistinguishable from
magic” feeling to some of what these tools appear to be able to do (until you look most closely or critically).

I see it as a very human problem, not a tech problem or even a tech issue.

We (collective “we”) tend to look past the flaws because these LLMs are very good at outputting what we expect them to output (Synchophancy as a Service).

Then it’s only one more mental leap to accept what the CEOs of these companies claim. (“AGI is imminent!”)

If you’re the CEO who makes GPUs, well you’d be leaving money on the table to NOT exploit these delusions.

If you’re any CEO with a website or app, you also don’t want to miss the bandwagon because the believers are so vocal about how this is “revolutionary” etc.

I don’t believe in conspiracies but I do believe in systemic, collective delusions of self interest.

[heybenji@social.coop]

@annaecook sunk cost fallacy.

[blazr@chaos.social]

@annaecook ignoring the reality is a widely accepted concept in the business of "leading" companies. CEO's just love it!

behaviour of the human being is mostly driven by continous irrationality. irrational behaviour is promoted by weird concepts/ideas like "I", "We" and "God".

[camerontw@social.coop]

@annaecook Finance Capital is not about accumulating money, but making money from the movement of money, so there needs to be something to spend investment money on, which in this case is building data centres; investors will get out before returns are relevant - image from https://www.patreon.com/posts/poetry-of-blade-43809667

[gatesvp@mstdn.ca]

@annaecook let's first talk about increased productivity.

Big white collar employers are in a big crunch right now. They need to deliver more of whatever it is they are delivering, but they have run out of employable humans and they have already deployed most of the modern technology that is available to make those humans faster.

This isn't just about infinite capitalist growth either. People need more and better access to the legal system, the medical system, the financial system, the education system.

So if you're a CEO of a white collar org, you are looking for the next big thing. You spent a decade and a half rolling out all the minor technical improvements, but there's nothing else you can do there. You need to find a real way to get more done with less so that people can actually afford to take their landlords to court, or get an education for a new job without drowning in a lifetime of debt.

So you go back to your tech vendors looking for the next hit... What have you got?... 🧵

[gatesvp@mstdn.ca]

@annaecook ...

And those technology vendors are like:

"We have this GenAI tool, it looks like it's really good at processing text as data, it might even be able to write software code. So you'll need less of those really expensive engineers that nobody can find."

You're the CEO of a bunch of professional paperwork pushers and now you have something that can push paper even faster? I mean, that sounds like an actual win.

And then, as icing on the cake, these tech companies are so desperate for people to roll this out, they offering it all to you at a discount. Possibly even a loss. Because you ran out of new tech last year, but they ran out of new tech in 2021.

So you sign the contract and you drop the mandate: "we're all in on AI, everyone is going to use it, we're going to be better". And soon everyone is doing it, nobody wants to get left behind, nobody wants to be the company that missed the best tech since the personal computer... 🧵️

[gatesvp@mstdn.ca]

@annaecook ...

why are companies continuing to go all in on it?

Because they don't have anything else.

They want more customers, they want more profits, they want more value in the world. And if they don't spend this money on AI technologies, what else are they going to spend it on?

I'm serious here. Instead of asking "why are companies spending all of this effort/money on AI?", flip it around, let's come up with non-AI things we think companies should do.

"Instead of spending money on AI, Company X should invest in Y".

Imagine you're the CEO of McKinsey, you push oodles of paperwork. You push oodles of paperwork, if you can't use an AI tool to help you process that paperwork, then what other investment should you be doing? Personally, I don't know. The high-speed paperwork interpreter sounds like a good investment. But I'm open ideas.

Rinse repeat for other industries: Insurance, Banking, Medical, Legal, Regulatory... if it's not AI, where do the research dollars go?... 🧵️

[gatesvp@mstdn.ca]

@annaecook

If AI does not

• have ROI when offered as a product / in platforms
• or increase productivity in employees

Here's the nuanced part. I don't think this is as true as we really make it out to be.

I have spoken to several small business owners for whom these tools have been incredible. It's letting them turn around customer requests in hours instead of days. It's letting them do analysis that would have been completely impossible without the robot assistance. I've know lots of software engineers whose day jobs are completely transformed by the technology.

Saying "it doesn't increase productivity" simply doesn't match with the people I've spoken with who are using this stuff for the very specific things it's good at.

BUT... it's not all roses... especially for big companies. And this is where you're seeing the most public failures and disappointments.

But those things are not problems with the technology, they're problems with big orgs, the technology is just surfacing them... 🧵️

[gatesvp@mstdn.ca]

@annaecook ...

This GenAI technology has some very specific limitations.

️ Converts most "text" sources into "data" sources
️ Transforms "data" between modalities
️ Create "data"
️ Create Stakeholder Consent

These things are incredible BS machines. They can generate BS all day at incredible speeds. But big companies were already drowning in BS, in bureaucracy. Big companies struggle to get Stakeholder Consent (also called "alignment").

Giving Big Orgs a BS machine doesn't solve their core problem. It doesn't give them the data they need to make big decisions. It doesn't give them Stakeholder Consent to select and prioritize projects. It doesn't make humans make faster decisions.

This is why all of the Big Orgs you see discussing this stuff are all reporting such dismal results. These tools can't help them fix their bottlenecks, it's just exposing those bottlenecks... 🧵️

[gatesvp@mstdn.ca]

@annaecook ...

And I've already seen this in the people I speak with. Software Engineers start cranking out features faster and suddenly the PMs can't keep up with specs. They have features available to ship faster than the Analytics team can complete significance testing on the effects of the change.

A lot of things we do still have a maximum human-imposed speed.

I have a friend who works on a Build/Release for a huge product. They ship Android and iOS very regularly. The company rolls out an "GenAI tools to increase productivity" mandate. They roll out some new tools, they're now getting 2x successful builds every day... they still only ship once / 2 weeks because iOS won't approve more than that.

The success of the GenAI tools could have been 100x, it wasn't going to ship the iOS app any faster. GenAI doesn't create Stakeholder Consent, Apple hasn't consented... 🧵️

[gatesvp@mstdn.ca]

@annaecook ...

The tools aren't (currently) helping to solve big corporate problems. The people I know getting the most mileage out of the GenAI tools are the ones using it for themselves. It's the people offloading parts of their own work onto the robots and simply getting more done faster. And then just kicking off work early or asking for promotions.

So big company programs initiatives aren't seeing the changes. Those research papers that show "no improvement from AI project X"... they're kind of meaningless.

The tools can help individuals because those white collar individuals are mostly data processing and transformation nodes and that's what the tool is good at. But big organizations are mostly decision making systems. Giant decision making graphs. And the tool doesn't make decisions... 🧵️

[gatesvp@mstdn.ca]

@annaecook ...

Stitching this all together, here are the short answers.

why are companies continuing to go all in on it?

They have nothing else to invest in.

If AI does not... have ROI when offered as a product / in platforms

It's not making decisions faster, so it's not improving the org. It doesn't matter how you bundle a shovel, it will not help you cut down a tree faster.

If AI does not... increase productivity in employees

I don't necessarily agree with this point because I have spoken with and read from too many people who have counter examples. But like I said, it could produce certain things at 100x the previous performance and the machine would still be moving at the same pace.

That performance boost can both be real and also be of no practical value to the company. And I think we're seeing a lot of that. 🧵️🪡️

(this probably needs to be a blog post somewhere) ️