Technology

    The information grey goo

    I’m broadly positive about the future of LLMs and AI, but no one should pretend there will not be difficulties or that the transition to using machines isn’t going to pose plenty of challenges. 

    Some scenarios, though, are profoundly dangerous, not just for the publishing and creative industries, but for society as a whole. 

    When we discuss the threat of AI, many people imagine rampant machine intelligences with big guns hunting us all down in a post-apocalyptic wasteland (thank you, James Cameron). I doubt that’s likely. But one consequence which I can see use sleepwalking into is the informational equivalent of an apocalypse that dates back over thirty years: the “grey goo” scenario.

    “Grey goo” was a concept which emerged when nanotechnology was the hot new thing. First put forward by Eric Drexler in his 1986 book The Engines of Creation, this is the idea that self-replicating nanobots could go out of control and consume all the resources on Earth, turning everything into a grey mass of nanomachines. 

    Few people worry about a nanotech apocalypse now, but arguably we should be worried about AI having a very similar effect on the internet. 

    Nowhere is safe

    Unless you haven’t been paying attention, you will have noticed that the amount of content created by LLMs has been increasing at a vast rate. No one knows how much content is being generated, but SEOs – whose job it is to understand content on the internet – are concerned. Less ethical SEOs have used a combination of scraping and generative AI to quickly create low-quality sites with tens of thousands of pages on them, reaping rewards in traffic from Google over the short term. 

    The problem for Google is that creating a site like that is the work of perhaps a week – and probably a lot less if it can be automated – while it takes months for the search engine to spot that it’s a low-quality site. With more automated approaches, it will become trivial to create spammy sites far faster than Google can combat them. It’s like a game of whack-a-mole, where there are moles appearing at an exponential rate. 

    And Google isn’t the only platform which AI is threatening to turn to mush. Amazon has a issue with fake reviews generated by AI. And although it claims it is working on solutions, it appears to be incapable of even spotting fake AI-generated product names.

    But what about human-to-human social networks? They have already been flooded with AI-generated responses. And it will only get worse, as companies create tools which let brands automatically respond to posts based on keywords using AI-generated text. Sooner or later, saying something which suggests you are in the market for a new car will get you spammed by responses from Ford, Skoda, VW, Tesla, every car dealer in your area, every private second hand seller… you get the picture. Good luck trying to find the real people. 

    It is obvious that anywhere content can be created will ultimately be flooded with AI-generated words and pictures. And the pace of this could accelerate over the coming years, as the tools to use LLMs programmatically become more complex. 

    For example, think about reviews on Amazon. It will be possible to create a programme which says “Find all my products on Amazon. Where the product rating drops below 5, add unique AI-generated reviews until the rating reaches 5 again. Continue monitoring this and adding reviews.” 

    We are already at the point where you can use natural language to create specialist GPTs. The ability to create these kinds of programmes is ultimately going to in the hands of everyone. And this applies to every rating system, all surveys, all polls, all user reviews – and similar approaches can be created for any kind of content. 

    Can Google, Amazon and the rest fight back? Yes – but at great cost. And it’s not clear that even the likes of Google has the resources to effectively fight millions of users of AI creating billions of low-quality pages at an accelerating scale.

    Model collapse

    A side-by-side comparison of content created from the same prompt in ChatGPT 3 versus ChatGPT 4 Turbo will show you the difference. And humans are getting better at writing prompts and giving AI models the information they need to do a better job. So surely, this is just a short-term problem, and AI content will get “good enough” to not flood the internet with crap.

    The issue is that there is a counterbalancing force at play. As more and more AI-generated content floods the public internet, more and more of that content will end up as training data for AI. Exacerbating this, quality publications are largely blocking AI bots, for entirely understandable reasons, which means less, and less higher-quality content is being used to train the next generation of models.

    For example, researchers have noted that the  LAION-5B dataset, used to train Stable Diffusion and many other models, already contains synthetic images created by earlier AI models. This is the equivalent of a child learning to draw solely by copying the images made by younger children – not a scenario which is likely to improve quality.

    In fact, researchers already have a name for the inevitable bad outcome: “model collapse”. In this case, the content generated by AI’s stops improving, and starts to get worse. 

    The Information Grey Goo

    This is the AI Grey Goo scenario: an internet choked with low-quality content, which never improves, where it is almost impossible to locate public reliable sources for information because the tools we have been able to rely on in the past – Google, social media – can never keep up with the scale of new content being created. Where the volume of content created overwhelms human or algorithmic abilities to sift through it quickly and find high-quality stuff.

    The social and political consequences of this are huge. We have grown so used to information abundance, the greatest gift of the internet, that having that disrupted would be a major upheaval for the whole of society.

    It would be a challenge for civic participation and democracy for citizens and activists, who would no longer be able to access online information, opinions, debates, or campaigns about social and political issues. 

    With reliable information locked behind paywalls, anyone unwilling or unable to pay will be faced with picking through a rubbish heap of disinformation, scams, and low-quality nonsense. 

    In 2022, talking about the retreat behind paywalls, Jeff Jarvis asked “when disinformation is free, how can we restrict quality information to the privileged who choose to afford it?” If the AI-driven information grey goo scenario comes to pass, things would be much, much worse.

    Apple's 27 per cent tithe

    9to5Mac:

    Apple has also confirmed that it will charge a commission on purchases made through alternative payment platforms. This commission will be 12% for developers who are a member of the App Store Small Business Program and 27% for other apps.

    The commission will apply to “purchases made within seven days after a user taps on an External Purchase Link and continues from the system disclosure sheet to an external website.”

    Apple had a chance to turn a legal defeat into a long-term victory. With Google charging 26% in the same circumstances, the company could have adopted rules which dramatically reduced the levy it wants to take, say to 12% for all developers. This would have gained the company a lot of credibility over the long term.

    But no. Instead, it chose to protect short-term revenue, and do something which looks petty, hostile to the developers who have made iOS a successful platform, and which will probably end up in court, again.

    When it comes to antitrust, perception matters. And sadly for them, in that area, Apple never misses an opportunity to miss an opportunity.

    The continuing challenge of return to office

    This is the first post-Substack edition of my newsletter, the first one delivered via WordPress. At some point in the coming days, the Substack version will be going away completely. If you want to know why, then you might want to read my post on Substack and platform risk, and then have a look at Platformer’s post on why it’s leaving Substack.

    It’s just over twenty years since the last time I worked completely in an office. In every job, I have had since 2003 when I left MacUser, I have spent at least a day a week working from home. And even before then, working away from the office was so frequent that I doubt there was a week between 1995-2003 when I wasn’t out for at least a day.

    Perhaps that’s one of the reasons I find the apparent desperation to get employees back into offices so strange. According to a survey by ResumeBuilder, 90% of employers plan to return to office by the end of 2024. Over a quarter of them intend to threaten employees who don’t want to return to office with being fired.

    But return to office hasn’t been plain sailing. The latest company to come a cropper with its plans is Internet Brands, the parent company of WebMD, which created a video “encouraging” its teams back into the office, ending in a screen – and I am not making this up – which notes “we mean business” and “don’t mess with us”.

    Cue much hilarity from the internet. Internet Brands is probably lucky that Twitter isn’t the force it once was because the pre-Musk social network would have run with this one for several weeks. Now, it’ll get buried under an avalanche of rubbish on Threads after a couple of days, or be seen by the couple of million users on Bluesky.

    But I digress.

    There are many arguments over the effectiveness of remote working vs in-office, some good and some bad. Having a separation between home and work is a good thing, for one, and although older employees tend to have plenty of space to make that work, younger ones in shared accommodation or living with parents often don’t. And younger employees also benefit from working closely with colleagues, who can mentor them informally much more easily when physically present.

    I think this is particularly true for the creative industries. Creative people typically like to think of themselves as the definitive solo fliers, coming up with great ideas and then hammering them into shape, before having them ruined by an editor.

    But the reality is that creative work is always collaborative. For example, in digital publishing, your content teams and audience development people need to mesh and work as one team; otherwise they can make the human equivalent of the sound gears make when grinding against each other.

    And that is why broad directives about the amount of time which people spend face-to-face vs remote are damaging not only to the people involved, who inevitably feel robbed of autonomy and disempowered, but also to a creative business.

    To find the most workable approach, leaders should focus on four factors: the needs of the work, the needs of the people, how work gets done, and the new managerial muscle required to manage a hybrid workforce.

    Broad, uniform directives can only be effective if all work is identical and performed by a uniform workforce, which is not the case in the creative industries, and not true for most others. Rather than imposing directives, CEOs should empower their managers, particularly those at the front line, to develop a comprehensive understanding of their team's work, the working styles of their team members, and the most effective ways to accomplish their tasks.

    Of course that depends on having empowered, well-trained leadership down to the small team level – but you are already doing that, aren’t you?

    Finally, A Grocery Cart That Can Save Me From The Horror Of An Ad-Free Moment Of Existence

    Finally, A Grocery Cart That Can Save Me From The Horror Of An Ad-Free Moment Of Existence | Defector:

    This is one of those new technologies that’s useful primarily as a viewfinder on a dismal present and a future determined to be even more miserable. Nobody anywhere will like the smart carts. Nobody, anywhere, will find them not-obnoxious. Everybody who does more than a couple of moments of thinking about it will be horrified by the idea of humanity digging gigantic devastating holes in the ailing planet and mining out its contents for the purpose of putting tablet computers onto grocery carts so that they can perform a service repulsive to literally everyone. Nobody—nobody nobody nobody!—wants to live in a society characterized by inescapable omnipresent advertising for consumer products; no one yet born has yearned to have video advertisements take up ever more of their field of vision.

    This is one of those paragraphs that I wish I had written

    Pluralistic “If buying isn’t owning, piracy isn’t stealing”

    Pluralistic: “If buying isn’t owning, piracy isn’t stealing” (08 Dec 2023) – Pluralistic: Daily links from Cory Doctorow:

    In Poland, a team of security researchers at the OhMyHack conference just presented their teardown of the anti-repair features in NEWAG Impuls locomotives. NEWAG boobytrapped their trains to try and detect if they’ve been independently serviced, and to respond to any unauthorized repairs by bricking themselves

    If you ever needed to see an example of quite how insane the “IP protection” laws are, this is probably it.

    Open extensions on Firefox for Android debut December 14 (but you can get a sneak peek today) | Mozilla Add-ons Community Blog

    Open extensions on Firefox for Android debut December 14 (but you can get a sneak peek today):

    Starting December 14, 2023, extensions marked as Android compatible on addons.mozilla.org (AMO) will be openly available to Firefox for Android users.

    But not of course for iOS, because Apple doesn’t allow companies to use any rendering engine other than Safari’s webview. And Apple also hates the idea of extensions that aren’t themselves applications, so don’t expect them to make the lives of extension developers easy once the EU forces them to open things up a little

    John G on Monica Chin's review of the Surface Laptop Go 3

    Daring Fireball: Monica Chin on the Microsoft Surface Laptop Go 3: ‘Why Does This Exist?':

    A $999 laptop that maxes out at 256 GB of storage and has a 1536 × 1024 display — yeah, I’m wondering why this exists in 2023, too. And I’m no longer wondering why Panos Panay left Microsoft for Amazon.

    The $999 MacBook Air has 256Gb of storage, 8Gb of RAM, and a three year old processor. I’m kind of wondering why that exists in 2023, too.

    Not to say that the Surface Laptop 3 is any good – it isn’t – but Microsoft isn’t the only company that has some bizarre pricing at the “low” end of its laptop range.

    Importing Apple Notes into Obsidian is now easy

    Obsidian’s Importer Plugin Lets You Move Your Apple Notes to Any Note-Taking App That Supports Markdown - MacStories:

    Apple Notes doesn’t have an export option. Instead, as Obsidian’s blog post on the Importer plugin update explains, it stores your notes in a local SQLite database. The format isn’t documented, but the developers of the plugin were able to reverse-engineer it to allow users to move notes and their attachments out of Notes and into two folders: one with Markdown versions of your notes and the other with the files attached to your notes. The folder with your notes includes subfolders that match any folders you set up in Notes, too.

    This is just outstanding work from the Obsidian team. There are a couple of limitations, mostly that it can’t import password protected notes (obviously), but I’ve tested it and it worked well.

    Related: undocumented SQLite databases should not be the way that a multi-gazillion dollar corporation is storing valuable data.

    Who would have thought Amazon would behave like this?

    Amazon deliberately deleted messages to hide dodgy business practices:

    The FTC also alleges that Amazon tried to impede its investigation into the company’s business practices. “Amazon executives systematically and intentionally deleted internal communications using the ‘disappearing message’ feature of the Signal messaging app. Amazon prejudicially destroyed more than two years’ worth of such communications—from June 2019 to at least early 2022—despite Plaintiffs’ instructing Amazon not to do so.”

    And the answer to the headline is, of course, “anyone that’s been paying attention.

    DOJ probing Tesla’s EV range cheating

    DOJ probing Tesla’s EV range after reports of exaggerated numbers - The Verge:

    The US Department of Justice (DOJ) is investigating the range of Tesla’s electric vehicles after reports surfaced that the company was relying on exaggerated numbers.

    In documents filed with the Securities and Exchange Commission, Tesla said that it had “received requests for information, including subpoenas from the DOJ, regarding certain matters associated with personal benefits, related parties, vehicle range and personnel decisions.”

    This follows on from a Reuters' report earlier this year, which found Tesla was getting so many complaints about range it was cancelling appointments with its service centres for customers with the problem:

    According to Reuters, there was nothing actually wrong with the vehicle’s battery. Rather, Tesla had allegedly created software to rig its driving range estimates to show a rosier picture. This led to thousands of customers seeking service appointments to figure out what was wrong with their vehicles. But because the vehicle was working as intended, Tesla’s diversion team simply canceled all the appointments.

    So Tesla created software which gave a false reading of battery range, then when people spotted it, they just canceled any service to them.

    It’s worth noting that when VW was caught cheating on its emissions tests by using a device to check when it was being tested and artificially improving results, it ended up being fined tens of billions of dollars.

    This isn’t on quite that scale, but regulators tend to take a very dim view of cheating customers. It’s quite possible this will cost Tesla billions.

    Anyone willing to bet that it will turn out this was done at Elon Musk’s insistence? And will that be the final nail in the coffin of his reputation?

    China launches investigation into iPhone maker Foxconn, says state media

    China launches investigation into iPhone maker Foxconn, says state media:

    China has launched an investigation into Apple iPhone maker Foxconn over tax and land use, Chinese state media reported on Sunday. The Global Times, citing anonymous sources, said tax authorities inspected Foxconn’s sites in the provinces of Guangdong and Jiangsu and natural resources officials had inspected sites in Henan and Hubei… The Global Times article quoted an expert saying “Taiwan-funded enterprises, including Foxconn . . . should also assume corresponding social responsibilities and play a positive role in promoting the peaceful development of cross-strait relations”.

    This is a very big deal and should be keeping Tim Cook awake at night. Effectively, it’s a small shot across the bows for Foxconn, a reminder that without the good graces of the Chinese government, it can’t exist.

    The new Apple Pencil

    Apple has released a new Pencil for iPad and it’s weird. It looks like the Second Generation Pencil (the one which charges by sticking to the side of the iPad Pro or current Air). And it will attach there. But it won’t charge if you do – it charges through a hidden USB-C port via a cable.

    Oh and it’s not pressure sensitive, which makes it worse for drawing than the old Pencil which charged via Lightning.

    It is, though, £79 rather than the ONE HUNDRED AND THIRTY NINE POUNDS the second generation Pencil will cost. So that’s one thing.

    Marc Andreessen's manifesto

    It would take a far, far longer post than I’m prepared to spend my time writing to go through Marc Andreessen’s “Techno-Optimist Manifesto” paragraph by awful paragraph, but a few points probably won’t go amiss. - If you’re going to approvingly paraphrase “a manifesto of a different time and place”, you might want to check that said manifesto’s author wasn’t an early member of Mussolini’s fascist party.

    - Writing “we believe technology is universalist. Technology doesn’t care about your ethnicity, race, religion, national origin, gender, sexuality, political views,” and then, two paragraphs later “We believe America and her allies should be strong and not weak” either shows you have no idea how to write, are being entirely disingenuous, or simply too stupid to think except in blocks of 240 characters. Either way, get an editor to help.

    - If you are going to talk about the Greek notion of arete then having an understanding of its relationship to class in Greek society might be a good idea, too. Aristocrats were assumed, by definition, to be exemplars of arete. It wasn’t something that thetes like me would have.

    - Believing that techno-optimism “is a material philosophy, not a political philosophy” while giving many repeated examples of what even a first year philosophy undergraduate which know was a political philosophy does not make you look smart.

    I could go on – the whole thing is riddled with howlers – but really is there much point?

    Thirty years ago, in a different life, I was a philosophy postgraduate student and taught first year undergraduates their introduction to metaphysics and ethics. In the first time, every time, someone would turn in an essay which read like this, and you would have to patiently explain to them they were going to have to rewrite it or fail, because philosophy does not mean writing down all the random thoughts you had when smoking that bundle of weed the night before the deadline.

    This is the manifesto of an emotionally insecure man having a mid-life crisis as he realises that his life’s work is meaningless and all the gold and treasure he has accumulated will never make him happy. Mid-life crises in men are often surprisingly redolent of the emotional outpouring of pseudo-intellectual silliness that accompany late teenage, that first period of life when boys start to realise they are not the centre of the world and lash out at the injustice of it all.

    Perhaps, then it’s no surprise this reads like it was written by a 14 year old and put on Pastebin. That it was written by a 52 year old with billions of dollars at his disposal says more about the failure of capitalism to imbue life with meaning than Andreessen could possibly imagine.

    EDIT: The first draft of this contained something about A16Z’s investment in Uber. In fact, they passed on Uber. But as if to make the point about the kind of technology which Andreessen believes will save the world as long as we never question it, let’s ask an AI...

    Screenshot 2023 10 17 at 08 48 51

    Publishers need to wake up to the truth about Google traffic

    Google-Extended does not stop Google Search Generative Experience from using your site’s content (searchengineland.com)

    Google explained that SGE is part of the Google Search experience; it is a search feature and thus it should work as how normal search directives work. “The context is that AI is built into Search, not bolted on, and integral to how Search functions, which is why robots.txt is the control to give web publishers the option to manage access to how their sites are crawled,” Google told us.

    I’ve been using both Bard and Bing CoPilot a lot lately and the direction is clear: while AI-driven search will link to original sources as references, they are not going to send much traffic your way. The aim is to provide the answer to any query on the results page, not one more click away.

    This has massive implications for publisher traffic, particular for reviews and answers pages which I think are most vulnerable to AI-driven answers. I’ve been using CoPilot for purchasing research and it’s great. I can start by asking it for, say, laptops under £1000 with good battery life. I can then have a conversation to interrogate more about each product. It’s a superior experience to any web page I have ever used for that kind of product research.

    Is it 100% accurate? No – but neither are a lot of reviews, particularly the kind of “best laptop for…” top tens that are written to hit the top of product searches on Google.

    But it’s not just affiliate: search provides between 40-80% of publisher site traffic. And we have already seen Facebook traffic, the other biggest referrer, die off.

    Publishers can no longer rely on Facebook and Google for the bulk of their traffic. The time has past when content strategies should focus on them. Instead, they need to focus on getting a loyal audience which they have direct relationships with. The SEO era is coming to an end, at least for large chunks of traffic.

    GitHub Copilot costs more per user than it charges

    Big Tech Struggles to Turn AI Hype Into Profits - WSJ:

    Individuals pay $10 a month for the AI assistant. In the first few months of this year, the company was losing on average more than $20 a month per user, according to a person familiar with the figures, who said some users were costing the company as much as $80 a month.

    The first stage of the enshittification cycle is often to charge customers less than it costs to run the service, in order to acquire and lock in as many as possible. After that, at some point, you dump on them from a great height.

    The good use of ChatGPT for factual writing

    I’m not a huge fan of using ChatGPT for writing, because even leaving aside issues of accuracy, its style is stilted and just the wrong side of formulaic. But there’s one area where it really works as a writing assistant: giving you an outline on a topic as a starting point.

    Tell it what you want and what to include, and it will come back with an outline of everything you should cover. It won’t be your final structure, but as a place to start and especially if you’re a bit stuck and need something to bounce around to fine-tune your idea, it’s a really good assistant.

    I’ve been thinking a lot about large language models as assistants for human creativity lately, in the context of Steve Jobs' old view of computers as “a bicycle for the mind” and also the Knowledge Navigator video which came later on – John Sculley’s vision of the future of computing. More to come on that…

    No, the UK government did not back down on its plans to spy on encrypted messages

    Many places reported that the British government had seen sense and backed down from its plans to require companies like Apple, Meta and Signal to give them back door access to end to end encrypted messages. Unfortunately, these reports were completely wrong.

    All that the government did was acknowledge that Ofcom, the body which would issue notices to companies requiring them to scan their networks, could only do so if it was technically possible – in other words, that it would be pointless to attempt to demand companies do something they physically couldn’t. This is clear in the quote from Lord Stephen Parkinson, the minister responsible, in the original FT story:

    “A notice can only be issued where technically feasible and where technology has been accredited as meeting minimum standards of accuracy in detecting only child sexual abuse and exploitation content,” he said.

    That is a long way from a government retreat. And as it stands, the clauses requiring companies to endanger their users privacy and security remain in the bill.

    The government has now confirmed this, with Michelle Donelan, the technology minister, saying saying:

    “We haven’t changed the bill at all… If there was a situation where the mitigations that the social media providers are taking are not enough, and if after further work with the regulator they still can’t demonstrate that they can meet the requirements within the bill, then the conversation about technology around encryption takes place,” she said.

    But the government is claiming it’s all technologically feasible:

    She said further work to develop the technology was needed, but added that government-funded research had shown it was possible. (My emphasis)

    The government is not backing down. It believes it’s possible technically, and will attempt to make companies comply. It will do this in secrecy, and it doesn’t give a damn about the privacy or security of the British people. The fight is not over.

    Elon Musk deliberately sabotaged a Ukrainian attack

    New Musk biography offers fresh details about the billionaire's Ukraine dilemma | CNN Politics:

    Elon Musk secretly ordered his engineers to turn off his company’s Starlink satellite communications network near the Crimean coast last year to disrupt a Ukrainian sneak attack on the Russian naval fleet, according to an excerpt adapted from Walter Isaacson’s new biography of the eccentric billionaire titled “Elon Musk.”

    As Ukrainian submarine drones strapped with explosives approached the Russian fleet, they “lost connectivity and washed ashore harmlessly,” Isaacson writes.

    How, exactly is this man not in prison? This is also quite telling:

    Gwynne Shotwell, Musk’s president at SpaceX, was livid at Musk’s reversal, according to Isaacson.

    “The Pentagon had a $145 million check ready to hand to me, literally,” Isaacson quotes Shotwell as saying. “Then Elon succumbed to the bullshit on Twitter and to the haters at the Pentagon who leaked the story.”

    Musk, like many stupid men, has been radicalised by Twitter into supporting the far right. That includes supporting Putin, the “white knight” who the far right thinks of as the saviour of western civilisation.

    Brexit making Windows worse in the UK

    Tom Warren, for The Verge:

    Microsoft will finally stop forcing Windows 11 users in Europe into Edge if they click a link from the Windows Widgets panel or from search results. The software giant has started testing the changes to Windows 11 in recent test builds of the operating system, but the changes are restricted to countries within the European Economic Area (EEA).

    “In the European Economic Area (EEA), Windows system components use the default browser to open links,” reads a change note from a Windows 11 test build released to Dev Channel testers last month. I asked Microsoft to comment on the changes and, in particular, why they’re only being applied to EU countries. Microsoft refused to comment.

    Of course this isn’t happening in the UK: thanks to Brexit, we’re not an EEA country, and so you’re stuck with Edge opening links from things like widget no matter what browser you choose. Now that’s what I call taking back control.

← Newer Posts Older Posts →