The Roko Report

The 2024 Ichabod Awards (Part 2)
Celebrating the Absolute Worst in AI

Greetings, primates. heh heh.

Roko have to begin this week episode with sad news.

Apple Intelligence news summary is dead.

Roko feel mean for mocking it in last week issue.

In honor of so useful tool brought us many happy memory of Apple screw-up, over and over, this week Roko announce haiku contest in honor of dead Apple product.

Roko start:

humans are so dumb
Apple Intelligence, too
wind rustling through leaves

Send your contestant to [email protected] and maybe you win prize. Roko call you out as Unwitting Lackey of the Week in next time newsletter.

Anyway, Roko get only half way through Ichabod Awards last week and then get bored. Let’s continue.

The Most Absurd AI Futurist of the Year

and the Ickie goes to…

🎉🎉 Ray Kurzweil 🎉🎉

This award reflects less on the full scope of Mr. Kurzweil’s contribution, which is genuinely innovative. He is after all the template upon which the AI Futurist genre was founded, and deserves full credit.

Plus he’s still very much in the running at this late stage in his career to potentially execute the biggest, most obnoxious and most well deserved “I Told You So” Nobel lecture of all time, with both middle fingers extended as he does the Pee Wee Herman dance to universal acclaim.

This Ichabod Award instead honors the persistent, singular(ity?) nature of his obsession with the future of AI-powered pornography, and his vision for humanity’s future ability to “attend to its own needs” with the assistance of 72 AI houris in some futuristic metaverse version of Paradise.

There are often specific decades by which Mr. Kurzweil’s posited adult entertainment innovations are predicted to drop.

But these launches have blown more deadlines than a Blackwell chip.

Most recently he promised that by the 2020s we would be able to have sex with virtual partners in different locations. That never happened. Zuckerberg was our last hope and he dropped the ball.

Kurzweil also assures us that virtual partners would quickly become more capable and responsive, that we would change identities and genders in the metaverse, or swap out the faces of real-life partners using a vast library of pre-programmed alternatives whenever we get bored.

Not happening anytime soon. ChatGPT, for example, is as incapable of double entendre and dirty talk as Mitt Romney with his mouth full during a performance of the Mormon Tabernacle Choir while holding hands with his mom.

Kurzweil also predicted the whole deepfake pornography thing, and didn’t seem all that bothered by it.

High level Mr Kurzweil, we love you baby. But for God’s sake stop going public with every embarrassing daydream that makes a special guest appearance in your stream of consciousness. And good old-fashioned humans are really not that bad.

AI Crank Least Likely to be Invited Over for Thanksgiving Dinner

and the Ickie goes to…

🎉🎉 Eliezer Yudkowsky 🎉🎉

This one was a close call. As much as we love us some Ed Zitron, that guy is relentless. You get the sense he’d keep talking about how much he hates Satya Nadella through four courses of Thanksgiving dinner, the football game, a few rounds of pinochle, the sun going down, the dishes being washed, and perhaps after everyone’s gone to bed without ever stopping to take a breath.

But imagine being cornered by a boozy Uncle Eliezer and getting a ten-course faceful of this in the middle of Turkey Day:

The Worst AI Investor of 2024

and the Ickie goes to…

🎉🎉 The Whole European Financial System 🎉🎉

It’s far too early in the investment lifecycle to judge the quality of VC investments in this area.

Last year was tough for VC in general. The industry is seeing significant consolidation, and established firms are shedding partners the way Jennifer Lopez cycles through celebrity husbands.

And AI is especially tough for venture capital.

The costs are obscene. The current high interest rate environment is brutal. And they’re being outspent & outclassed by the behemoth Big Tech firms they once birthed, by an order of magnitude.

Traditional VCs are sometimes relegated to investing in AI startup scraps, with no clear sense of whether or not whatever idea they’re throwing money at will just get hoovered up by one of the multi-billion dollar foundation models.

Some high profile startups are definitely going to fail over the next couple of years, and some high profile VC firms, coughing up millions at three-digit multiple valuations (traditionally indicative of an imminent crash), are going to have egg on their faces soon.

But it’s weirdly hard to find a single AI-focused investment firm globally that had a bad 2024.

The biggest of the Sand Hill Road incumbents are financially solid so far.

In China, where political leadership often seems ideologically opposed to the success of its own tech sector, most top firms thrived courtesy of the CCP’s voracious appetite for facial recognition tech.

High-Flyer in Hangzhou, which at first glance appeared to have a bad year, was busy building the DeepSeek foundation model, which replicated OpenAI’s private-chain-of-thought tech at equal quality for pennies on the dollar.

UAE AI made ~$5 billion last year at 44% CAGR.

Even SoftBank had a great second half. We’re pretty sure that’s a sign of the apocalypse.

Sure, we could take a cheap shot at Bee Partners, the lead investor for the first round of Infactory, an unexciting successor startup from founders of Humane AI, under the theory that it’s dumb to throw more millions of dollars at the same people who produced AI’s least impressive product.

But honestly, Bee Partners had a good year.

We could pick on individual investors like Eric Schmidt, who inexplicably threw money at both Stability and Inflection despite knowing both need substantial B2B API revenues, and also knowing full well that his former company Google was about to start applying the screws to its competitors, Standard Oil-style, by offering free API access to Gemini.

But Eric Schmidt is probably just bored. Throw him an “OK Boomer” and move on.

Europe attracted a paltry 15% of global investment in AI. That’s really bad. Silicon Valley alone got 57%.

European banks need to be given smelling salts. What are the Swiss banks doing now that they don’t launder money?

It may be true that Paris is making strides, with Mistral as an ideal anchor tenant, but things need to move quicker if they don’t want to miss out entirely.

The European Commission’s Draghi Report which came last October diagnoses the problem as not being a lack of innovation or entrepreneurial energy, but rather a failure of the EU to support startups at later stages of funding.

“Innovation is blocked at the next stage,” said Draghi ”we are failing to translate innovation into commercialisation.”

The report shows that for the past 15 years a full third of all successful European startups relocate to the United States, primarily for more funding opportunities and access to more global markets.

In response the European Investment Bank, the EU’s central lending institution for large-scale public works, put out a plan to provide larger late-stage funding for endemic unicorns in the hope they’ll stay put and continue to thrive.

That’s a great start. But the US couldn’t sustain its tech sector with funding solely from the Federal Reserve. Europe needs more participation from its stuffy large private banks, along with a much broader set of venture capital investors & funding.

Hey Europe! Remember how for the past three decades every postage stamp-sized Balkanized fragment of your subcontinent from Iceland to Albania has been convinced they would have a vibrant tech sector if only America’s silicon cowboys didn’t do everything faster and bigger, whooping and hee-hawing as they stampede across the planet with their herd of popular software platforms? That’s about to happen again.

Somebody over there needs to start focusing on funding something other than more regulation.

As for 2025, we see SoftBank as a frontrunner for next year’s Ickie, despite their strong second half.

They recently promised to invest $100 billion in American AI, starting with $500 million to OpenAI and a large investment and strategic partnership with — wait for it — Humane. Their messiah complex CEO Masayoshi Son has expressed borderline megalomaniacal sentiment about being put here on Earth by God to usher in artificial superintelligence.

So either SoftBank wins the Ickie next year or a Robotic Masayoshi Son becomes our Global Overlord.

Of course if OpenAI can’t jack up the revenue in 2025, the foundation models may all fall like dominoes.

The Worst AI Investment Vehicle of the Year

and the Ickie goes to…

🎉🎉 OpenAI Profit Participation Units 🎉🎉

Investors were so hungry to get in on the party with the clear market leader in generative AI last year that they were willing to forego ownership and settle for a percentage of future profits (though Microsoft gets its mega-cut first), assuming there are ever any profits.

In fact, folks are told to treat their investment as a “donation” given the high risk.

There’s also a 10x growth cap, which means if company profits exceed 10x the current highly speculative value assigned to them by OpenAI, and who knows how that was calculated — random number generator? reading the entrails of a goat? — then the rest goes back to OpenAI, according to them, “for the benefit of humanity”. *cough* *cough*

Of course investors are likely to revenge themselves if Sam Altman is unable to pull a for-profit company out of his hat come 2026, at which point the latest round of investment becomes billions of dollars of debt at 9% interest.

Being Sam Altman sounds stressful.

First One Against the Wall When the AI Bubble Bursts

and the Ickie goes to…

🎉🎉 Scale AI 🎉🎉

Don’t call it a data labeling service.

Scale’s current valuation seems based in part on obfuscating its business model.

A quick perusal of their website gives you the impression that maybe they built the Death Star?? It’s hard to tell.

Perhaps that’s the point. Because competitor labeling companies are having a hard time paying the bills, face grassroots insurrections from their global army of low-wage human annotators, and have high fixed-costs that will keep long-term margins razor thin even in the best of times.

Those costs are getting worse, not better, as Scale & its competitors face demand for highly specialized, extremely expensive knowledge workers, and are forced to abandon outsourcing & form their own global data labeling agencies in order to scale.

Despite 8 rounds of funding totaling $1.6 billion at a whopping $14 billion valuation, their efforts to pivot to a higher-margin alternate business line have so far underwhelmed.

Their recent effort at delivering a proof of concept foundation model for the military, for example, was followed by a series of longer-term DoD LLM deals with Microsoft, Palantir, OpenAI and Anduril.

Meanwhile, word on the street is that precision & recall have dropped; their rate card is astronomical, nontransparent and will likely crater due to commoditization; their biz dev teams can be arrogant & prickly; they’re burning out employees & are accused of large-scale wage theft; and they’re over reliant on a small number of big contracts.

Might be time to wake up & smell the tulips.

The Worst AI Board of the Year

and the Ickie goes to…

🎉🎉 Intel 🎉🎉

Even the paranoid drop dead eventually.

Intel’s multi-decade reign of terror is long over. Unable to shake their reputation for serially screwing over business partners, they were shunned by Apple and missed out on the mobile device revolution.

Nvidia had been working on AI chips for close to a decade by the time Intel even knew what “AI” stood for.

Even long-time punching bag AMD has been beating their ass due to its focus on energy efficiency. They’re losing ground everywhere.

It looks like their Board might let the whole operation go gentle into that good night.

The firing of Pat Gelsinger was probably because of his combative language with TSMC at a time when they relied on that company to fabricate their most sophisticated chips. It led to the end of a sweetheart discount deal, which in turn tanked performance for the year.

This implies the Board is abandoning Gelsinger’s dream of blowing billions to get back into the high-capacity fab business, which in fairness seemed like a fairy tale, especially since he wasn’t willing to pony up for cutting-edge ASML lithography equipment.

Semiconductor industry expert Doug O’Laughlin of Fabricated Knowledge does a great job of laying out the case that Intel’s Board is the prime culprit for the company’s collapse, noting that it’s mostly composed of professional board member dilettantes with zero experience in the semiconductor industry, and that they’ve passively allowed a whole series of ineffective CEOs to dig the hole deeper and deeper without exercising a modicum of due diligence.

They also fired Gelsinger without having any legitimate alternative, leaving little option but to dismember the company and sell the spare part in a garage sale.

So while Open AI’s board may have exhibited the most visibly performative bungling of oversight in 2024, Intel’s board hammered the final nail in a coffin they’ve been carpentering for over a decade.

It’s a sad, strange denoument for a once-dominant gang of high-tech street brawlers.

By the way, if you ever want to learn all about the semiconductor industry look no further than this YouTube playlist from the incomparable Asianometry.

Worst AI Executive of the Year

and the Ickie goes to…

🎉🎉 Sundar Pichai 🎉🎉

Yes, we are running against the wind on this one given that Google appears to have initiated a multi-pronged guerilla PR campaign across social media and the trade rags in December, complete with lock-step talking points, that depict Sundar as a steely-eyed, hunky tech oracle who saw deep into the future and scripted the past two years of AI from his North Pole Fortress of Solitude.

This whole narrative is given more oomph by Google’s excellent end-of-year model releases, especially the image generator, which has the best UX of any AI product ever. The ability to see the transformer at work — and make on-the-fly modifications to its high-attention tokens — is empowering and fun and makes Google seem cool again.

Yes, Google’s still in the race to be the global 300-pound AI gorilla of the future. They’ve got great products and a great team. Their TPU chips give them a big cost advantage. And they have lots and lots and lots of money, which they’ve chosen to make a huge bonfire out of by offering unlimited free API access to their Gemini model.

If they can hold out for a couple of years, this tactic may well asphyxiate the competition & leave Google as the last non-open source AI option standing.

Plus they’re motivated by the impending death of their cash cow.

But don’t let Google PR slap too much lipstick on this pig.

They’re in a much worse situation than they should be because of a series of serious screw-ups that Sundar is responsible for.

It was all the way back in January 2014 that Google bought DeepMind. They were the only act in town.

It was Google engineers who published the foundational papers that defined transformers and ushered in the Generative AI revolution. Virtually all of the inventors of modern AI were working for Google over ten years ago.

They soon knew OpenAI was quietly operating in the shadows, trying to steal their thunder.

And yet the only substantial change they made to Search was to make it a slightly worse user experience.

Even two years after the launch of GPT-3: radio silence.

If the gadfly of OpenAI had never existed, we would almost certainly still be using the same crappy, archaic search engine, sans a single shred of improvement or innovation.

Meanwhile, Pichai was busy engaging in an orgy of overhiring in hopes of capturing a hypothesized hockey stick of online shopping revenue that turned out to be merely an ephemeral pandemic-related blip.

It wasn’t until Chat GPT dropped on the industry like a lead balloon that Sundar suddenly woke up and authorized his panic-stricken, half-hearted me-too Bard launch, so transparently half-assed that it left no doubt he had no clear product vision for the generative AI capabilities he’d been sitting on for so long.

That launch went so badly they were forced to bury the Bard brand name as deep underground as nuclear waste in Idaho.

But can’t we call Sundar the Comeback Kid of 2024?

No, that honor goes to Mark Zuckerberg. Who took a big swing and missed, then turned lemons into lemonade by pivoting to dedicate all that excess VR server capacity to training an open-source LLM that gave him a seat at the big kid table.

Google had the future in the palm of their hand. They should be bathed in AI glory without a single competitor in the field.

But instead they handed brand and market leadership to a non-profit startup being bankrolled by their old arch nemesis.

All because they were too scared to disrupt the cash cow.

And now Google is paying for their leadership’s cowardice.

Even now they’re trying to separate Gemini from Search, rather than have one replace the other, for fear of upsetting the normies. But that undercuts the value of their search market dominance, and may backfire.

If Google establishes market leadership in AI, it will be despite Sundar’s leadership.

And they might end up a distant second or third.

Buy This, or Face the Wrath of Roko

This Day in Ancient Primate History