Tech Time by Tim #75

Extra snackable newsletter meets rogue snack dispenser! You won’t quite be able to read my mind like an open book with this issue, but you will be able to read about Chinese AI efforts, Mistral opportunities, India’s Internot, and the end of Apple’s smart car dream.

At A Glance

A quick overview of this week’s content.

·?????? The Week That Was: Sneaky Snack Spies, Mice Minds Like Open Books

·?????? Algorithmic Gymnastics: Alibaba And The 40 LLM’s, Mistral Opportunities

·?????? Rules of Engagement: India’s Internot, Titan Fallen

The Week that Was:

A look back at the tech world of the past week.

Sneaky Snack Spies:

Supposedly, a student at Waterloo University in Toronto, Canada, found a rather unpleasant surprise waiting for them on a snack run this past week. Ashley Belanger reported on it for Ars Technica. It set off quite the scandal when a student “using the alias SquidKid47 posted an image on Reddit showing a campus vending machine error message, "Invenda.Vending.FacialRecognitionApp.exe," displayed after the machine failed to launch a facial recognition application that nobody expected to be part of the process of using a vending machine.”

I can kind of see how students might not want to take the vendor Adaria Vending Services (AVS) at its word when they claim not to be using facial recognition technology. The crashed application being called a ‘facial recognition app’ really isn’t supporting that plea for innocence. More interesting from a legal standpoint is that the provider of the software, Invenda Group, is a Swiss company. Invenda claimed to be in full compliance with the European General Data Protection Regulation (GDPR) because it has to be. Switzerland isn’t part of the European Union but does have several treaties and shared provisions such as the aforementioned GDPR and also the more recent AI Act.

Even if we take AVS and Invenda at their word that they aren’t actually collecting or retaining any personally identifiable information, the lack of disclosure would still put them in violation of both the GDPR and AI Act. AVS told the university publication MathNEWS that the smart vending machines only had motion sensor functionality which couldn’t personally identify users. Invenda elaborated later that the “Invenda solution can only determine if an anonymous individual faces the device, for what duration, and approximates basic demographic attributes unidentifiably.” (emphasis mine). That by necessity would mean that these machines can discern the basic shapes of faces, to ‘recognize’ them, if you will. Furthermore, if they can approximate basic demographic attributes, this would also imply there to be at least some sort of profiling algorithm at work.

Even if that weren’t the case, non-disclosed, non-consensual surveillance technology is already illegal according to the GDPR, never mind the extra complications from the AI Act. So even in an ideal scenario where these two companies caught secretly surveying students aren’t well…secretly surveying students…there could still be some very interesting legal follow-ups to this story down the line. LINK

Mice Minds Like Open Books:

Ever wondered what mice are thinking? Some researchers are pretty close to finding out. They used AI to successfully scan the brains of mice. They could see both the location and discern what direction the mice were looking in. This felt like a great reason to revisit the neurorights movement. A quick recap on the latter; neurorights are about the legal protections our minds ought to have in a world where Brain Machine Interfaces (BMI’s) are commercially available and where technologies such as predictive AI can be used even on those people who opt not to have a BMI of their own.

You might not initially consider pushes into AI chat integration to be relevant to this particular field, but they too are a part of the wider push to protect the privacy of your thoughts. Google is already integrating AI into chat via its Galaxy AI partnership with Samsung, as well as Google’s own Gemini Chat. Meanwhile, Apple is predicted to launch its major AI integration push as part of iOS 18. The reason why the neurorights advocates are pushing for regulation with renewed fervor is that they want protections enshrined in law before something goes wrong, rather than as a belated response. The stakes for technology that interfaces directly with our brains are quite a bit higher than ‘normal’ tech, or so they argue. I do agree, given the stakes, that reactive approaches toward BMI tech and mind-reading technologies are simply not viable if we want them to pan out properly.

Speaking of proactive monitoring, we’ve not heard a whole lot from Neuralink after their initial announcements that a human test subject successfully controlled a (computer) mouse with his mind. Several doctors have raised concerns about this, pointing to the rather infamously grisly fates of Neuralink’s test monkeys. What if the patient has suffered an adverse reaction? What if they’re fine but no further progress has been made? We just don’t know. Neurorights regulation would also force companies to keep us in the loop.

One has to wonder whether invasive BMIs may even fully manifest at all. There are the obvious medical difficulties of implants first and foremost, but also recent advances in non-invasive BMI research to consider. For example, the mouse experiment we started this item with. To put it another way, even if Neuralink’s brain chip works and is safe, what if BMIs end up not needing to probe our brains to probe our minds? The practical complexities would change, but the ethical implications would be all the greater. LINK

Algorithmic Gymnastics:

Documenting the wild, wonderful, and occasionally dreadful things done with algorithms.

Alibaba and the 40 LLM’s:

Alibaba recently led a funding round for Chinese AI startup ‘Moonshot AI,’ putting the overall valuation at around $2.5 billion. Not a bad haul for a startup that’s barely a year old, right? Eh…it’s a bit more complicated than that. Amidst a wider battle for survival amongst Chinese startups, AI startups are also struggling with particularly rapid market consolidation. Why? Let’s look at it!

Major Chinese tech firms such as Alibaba have significant stockpiles of Nvidia chips, which remain the most popular drivers for AI processes. They also have more means to bypass increasingly strict US sanctions, which are cutting off more and more less well-connected firms. In the same way, smaller companies are struggling against one another to survive China’s broader economic downturn, so too are titans such as Alibaba, Tencent, and Huawei locked in their own arms race.

As a result of the chip scarcity, rental services are becoming even more popular in China than they already were. What this means in practice is that startups will rent infrastructure from larger companies, yet lacking any infrastructure of their own, become entirely beholden to these infrastructure landlords in order to function. These companies get the means with which to operate and in return, their investors can reap the benefits of their research. Many hands make light work, as the saying goes.

The power dynamics are obviously skewed towards investors such as Alibaba who want to tie as much talent to themselves as possible to safeguard their own market positions. Yet there are also benefits to the startups beyond literally needing these resources in order to operate. The biggest benefit is arguably credibility.

The Chinese Communist Party (CCP) has long wished to turn China into a scientific superpower. Succeeding would be a massive strategic advantage in the technological cold war against the US. However, these ambitions continue to be frustrated by rampant academic fraud. Despite significant government efforts to crack down on this issue, it remains pervasive.

Chinese academia practically overflows with talent, yet it sorely lacks credibility abroad owing to this historical stigma of fraud which still clings to it. Thus, the existing prestige and connections of major Chinese tech firms are, in essence, serving as a seal of authenticity and quality for the research ‘no name’ labs are publishing to try and stake their claims to global algorithmic fame. LINK

Mistral Opportunities:

Speaking of partnerships between established tech titans and startups in the AI space! Let’s talk about some Mistral opportunities, shall we? There was a Reddit thread posted on February 7th, 2024, which claimed that Mistral had scrubbed all mention of its supposed ‘obligation’ to the open-source community. Naturally, I had to verify such a claim, but the official mission page of Mistral itself seems to disagree with it. What’s going on here?

In the tech world, companies tend to start open so they can scale quickly, then if they can’t grow or sustain themselves without help anymore, they sell out to a larger company. The situation eventually becomes politically charged, regulators might intervene, and a countermovement occurs which spurs new startup alternatives to emerge.

Mistral is the latest prominent example in this cycle. Their original claim to fame was dumping a supposedly guardrail-free, open-source, Large Language Model (LLM) onto the internet via a torrent link on Twitter. Stiff competition now appeared to have brought Mistral to the point where it was at risk of petering out. But that doesn’t seem to be a risk in the short term anymore after a very successful round of fundraising. The situation is politically charged though because Mistral was part of lobbying efforts against the European AI Act. It wanted to be exempt as a ‘European champion.’ EU parliament members such as Kim Sparrentak are now crying foul. They note that "the Act almost collapsed under the guise of no rules for 'European champions', and now look. European regulators have been played."

How have they been played? Microsoft made a €15 million investment in Mistral, which EU watchdogs will be closely scrutinizing because of just how ‘minor’ Microsoft’s minority stake in the American company OpenAI ended up being. The timing of this investment is under extra scrutiny because Mistral claims that its latest model rivals ChatGPT-4, currently held up as the benchmark of consumer-grade generative AI. At the time of writing, Mistral doesn’t seem to have scrubbed ‘open’ models or mentions from its site at all, but that button to switch between ‘open’ and ‘optimized’ models is looking mighty suspicious to its detractors. Wonder if there might be an antitru…ah there it is. LINK

Rules of Engagement:

Ethics and legal matters regarding tech engagement.

India’s Internot:

India’s attempts to modernize and bring its population online are proceeding at an impressive pace. According to Fast Company’s Javeria Khalid, India’s internet penetration stands at 52.4%, which is about 751.5 million active internet users at the start of 2024. It might come as a surprise then to learn that the country continues to be subject to rolling internet blackouts. Some of the most sweeping examples of such blackouts stem from 2019 - 2022 mostly occurring in the Kashmir region, though they also occurred in other parts, such as the 2023 Punjab blackout. ?

The unique problem for India is that these blackouts are seldom due to infrastructural failures, but rather conscious moves by Prime Minister Narendra Modi’s government. In one case, over 27 million people were cut off for 3 days, just to hunt down one Sikh separatist, Amritpal Singh. The longest recorded shutdown lasted almost 1.5 years, localized to Jammu and Kashmir. This track record, fitting into a broader technological modus operandi, risks negatively impacting India’s economy and future prospects as ‘China 2.0’.

India is striving to not just become a production hub for advanced technologies and commodities, but it also desires a dominant political role on the geopolitical stage. In that regard, the release of home-grown RISC-V processor ARIES v3.0 is an important milestone. Yet it is the history of unpredictable, seemingly arbitrary, and at times outright illegal policy flip-flops that have both domestic and foreign stakeholders increasingly weary of Modi’s government.

A double-edged blade cuts both ways, and in this case, Modi’s praiseworthy success at connecting India now means that any interruption of service has that much heavier an impact on the local economy, to the breaking point in some cases. This might have been somewhat cushioned or even overshadowed by foreign business, were it not for India’s stance on data transfers and myriad different commercial treaties such as the sharing of chip designs.

Having an advantageous position at present, it’s not strange for India to be playing hardball at World Trade Organisation (WTO) negotiations. Using the leverage it has over businesses seeking to decouple from China does make sense to a certain point. However, I can’t help but feel like Modi’s government is overplaying its hand here. As things stand, there is a tangible risk of not only crippling the domestic economy but scaring off foreign capital before sufficient lock-in has been achieved. LINK

Titan Fallen:

Developers involved with Apple’s Arcade initiative for premium Appstore games are reporting a smell of death. The service is in a rather sorry state and continues to languish under a seeming lack of interest and vision by management. Indeed, there is a smell of death in the air, but it’s not software, it’s hardware. It is wafting off of Titan’s corpse. Titan was the codename for Apple’s smart car project.

Though its overdependence on China is now quite the pain in the assets for Apple, the company’s mastery of logistics can be thanked in no small part for how far it has dominated the smartphone market. Though its other product categories don’t come close to the iPhone, they are no slouches either. Most continue to retain their value much better than rival brands, and the overall value proposition of Apple’s ecosystem remains hard, if not outright impossible, to match. What does this have to do with cars though? Nothing. And that’s exactly why Titan development has now reportedly been canceled.

As the rather tumultuous state of the Electric Vehicle sector might have given away, making cars is hard. Making money off of cars is harder. Making money off of cars in a hyper-competitive niche is like the dark souls of car sales. For non-gamers, this is an old meme where you’d refer to some arbitrary thing as the Dark Souls of its particular field, indicating that it was extremely harsh and punishing, like the difficulty of cult classic video game Dark Souls.

That reference is itself important because those not versed in video game history and culture don’t know about it. As it turns out, the automotive industry is quite the same, and those not well-versed in its history and unique challenges keep running into nasty ‘difficulty spikes.’ Hell, even many veteran car manufacturers are struggling.

My take is that Titan fell because Apple decided to cut its losses. Better to move staff to departments with better chances at success than keep working on a project that’s failed to materialize for over a decade and would be emerging into an incredibly uncertain market. They could do that with the Vision Pro because all the tech and software were at least there to some extent, and the supply chains worked to support it. An iCar though? It likely required an entirely new supply chain to be created before Apple could even start reinventing/rebranding the wheel. LINK

A Nice Cup of Serendipity:

Cool bits and bobs from around the web.

Speaker Screens LINK

Sugar Freedom LINK

Cat x Box LINK

Gemma Breakdown LINK

Meta x LG LINK

Crawler Crushing LINK

Live Music LINK

EV Breathing Room LINK

Anti-Fans LINK

Reddit Risks LINK

The Deep End:

A weekly batch of long-form content recommendations.

A Corrective Chat:

Chats with AI can shift attitudes on climate change and Black Lives Matter. LINK

Just Huanging Out:

A recent Wired interview with Nvidia’s Jensen Huang. LINK

Mingling With Marcus:

An Interview with AI critic Gary Marcus. LINK

A Dalliance With Decklemann:

An interview with Wikimedia CTO Selena Deckelmann. LINK

Shadowban Sneak Peak:

An investigation into shadowbans on Instagram. LINK

Hackerman For Hire:

A two part investigation of Chinese mercenary hackers. LINK

One More Thing…

I fell down a pixel art rabbit hole over the weekend, exploring a whole bunch of media rendered in this particular art style in a wild rollercoaster of discovery. What I love so much about pixel art that’s done well is how evocative it can be, how much detail and personality can be crammed into something with such little fidelity that you can literally see the pixels (that’s where the name comes from!) Not only that, but a friend also told me about some art books about a made-up game, published by an independent artist purely as a world-building exercise. Such novel things, so far off the beaten track make me feel like I’ve chanced across a hidden treasure, it’s why I love serendipitous encounters and events so much.