I've been in Firefox a long time

I intended to write a "history of my browser usage" post as part of a longer piece on the Chrome monoculture, but this blog is nearly 20 years old and it turns out I already did a history.

I can't find when I permanently dropped SeaMonkey, but it was after 2010. I dropped SeaMonkey late 2013 (thank you stale profile directory with a date stamp) when it was clearly abandonware and I learned you actually could launch Firefox in parallel with multiple profiles (the

firefox -p --no-remote

combination was key). I stopped using multiple profiles when the Container plugin came out that did nearly everything separate profiles did. It turns out SeaMonkey is still getting updates, but it seems to be tracking the Firefox and Thunderbird Extended Service Releases.

For those of you too old to remember the original Netscape Navigator, it also came with a few tools beyond the browser:

  • The browser, of course
  • An email client, since this was before GMail and web-editors for email weren't really a thing yet
  • An HTML editor (for pre-CSS versions of editor)

The reason I liked SeaMonkey and Opera is they both still shipped with an email client. It was pretty nice, actually. I kept Opera around as my email client way past when I stopped using it for general browsing. I'm fuzzy on what I did after Opera dropped their mail client, I may have grumpily transitioned onto Gnome Evolution at that point. Also, Gmail was out and I was quite used to web-based email clients.

So yeah, I've been in Firefox for over a decade at this point.

This is a controversial take, but the phrase "it's industry standard" is over-used in technical design discussions of the internal variety.

Yes, there are some actual full up standards. Things like RFCs and ISO-standards are actual standards. There are open standards that are widely adopted, like OpenTelemetry and the Cloud Native Computing Foundation suite, but these are not yet industry standards. The phrase "industry standard" implies consensus, agreement, a uniform way of working in a specific area.

Have you seen the tech industry? Really seen it? It is utterly vast. The same industry includes such categories as:

  • Large software as a service providers like Salesforce and Outlook.com
  • Medium software as a service providers like Box.com and Dr. Chrono
  • Small software as a service providers like every bay area startup made in the last five years
  • Large embedded systems design like the entire automotive industry
  • Highly regulated industries like Health Care and Finance, where how you operate is strongly influenced by the government and similar non-tech organizations
  • The IT departments at all of the above, which is much smaller than they used to be due to the SaaS revolution, but still exist
  • Scientific computing for things like space probes, satellite base systems, and remote sensing platforms floating the oceans
  • Internal services work at companies that don't sell technology, places like UPS, Maersk, Target, and Orange County California.

The only thing the above have any kind of consensus on is "IP-based networking is better than the alternatives," and even that is a bit fragile. Such out there statements like "HTTP is a standard transport" are ones you'd think there would be consensus on, but you'd be wrong. Saying that "kubernetes patterns are industry standard" is a statement of desire, not a statement of fact.

Thing is, the Sysadmin community used this mechanic for self-policing for literal decades. Any time someone comes to the community with a problem, it has to pass a "best practices" smell test before we consider answering the question as asked; otherwise, we'll interrogate the bad decisions that lead to this being a problem in the first place. This mechanic is 100% why ServerFault has a "reasonable business practices" close reason:

Questions should demonstrate reasonable information technology management practices. Questions that relate to unsupported hardware or software platforms or unmaintained environments may not be suitable for Server Fault.

Who sets the "best practices" for the sysadmin community? It's a group consensus of the long time members, which is slightly different between each community. There are no RFCs. There are no ISO standards. The closest we get is ITIL, the IT Infrastructure Library, which we all love to criticize anyway.

Best practices, which is "industry standard" by an older name, have always been an "I know it when I see it" thing. A tool used by industry elders to shame juniors into changing habits. Don't talk to me until you level up to the base norms of our industry, pleeb; and never mind that those norms are not canonicalized anywhere outside of my head.

This is why the phrase "it's industry standard" should not be used in internal technical design conversations

This phrase is shame based policing of concepts. If something is actually a standard, people should be able to look it up and see the history of why we do it this way.

Maybe the "industry" part of that statement is actually relevant; if that's the case, say so.

  • All of the base technology our market segment run on is made by three companies, so we do what they require.
  • Our industry are startups founded in 2010-2015 by ex-Googlers, so our standard is what Google did then.
  • Our industry computerized in the 1960s and has consumers in high tech and high poverty areas, so we need to keep decades of backwards compatibility.
  • Our industry is VC-funded SaaS startups founded after 2018 in the United States, who haven''t exited yet. So we need to stay on top of the latest innovations to ensure our funding rounds are successful.
  • Our industry is dominated by on-prem Java shops, so we have to be Java as well in order to sell into this market.

These are useful, important constraints and context for people to know. The vague phrase "industry standard" does not communicate context or constraints beyond, "your solution is bad, and you should feel bad for suggesting it." Shame is not how we maintain generative cultures.

It's time to drop "it's industry standard" from regular use.

Some engineers at Google have put forth a proposal called Web-Environment-Integrity that has the open source community up in arms. The leading criticisms of this proposal are "Google wants to make DRM for websites" and "Google wants to ban ad-blockers." These are catchy headlines intended to capture attention, they're also mostly true. For the people who don't want to wade through the discourse, this post is about what WEI does and where it came from.

This story begins in the previous decade when Google put forth the "Zero Trust framework" as a way to get rid of the corporate VPN. Zero Trust was a suite of techniques to allow companies to do away with the expensive and annoying to maintain VPN. The core concept behind Zero Trust was something I didn't truly understand until a few years ago: Zero Trust adds device attestation (a machine login) in addition to the user attestation when deciding whether to grant access to a resource, which is a more robust security boundary than a separate VPN login.

In a company context, you can reasonably expect the company to own both the server and the machine employees are accessing internal resources from. Zero Trust also enabled servers to specify a minimum security level that clients must meet in order to have access granted, such as up to date browser and OS versions, as well as a valid device identifier. This ends up being a major security win because when an employee has their user credentials phished, an attacker can't immediately use those stolen credentials to do evil things; the attacker will have to somehow gain a device identity as well.

In companies that use VPNs, phishing the user's credential was often enough to allow creating a VPN connection. If that wasn't enough, Phishing the VPN certificates would do the trick. Full Zero Trust makes the attacker's job harder. Again, in a corporate context where the company reasonably owns both the client and server sides of the conversation.

Back to WEI: Web-Environment-Integrity is a proposal to bring a limited form of device attestation to public surfaces. While WEI won't have a device credential, it will have the ability to attest to OS and Browser versions, presence and state of browser plugins, among other things. In theory this allows bank website operators (for example) to ensure people have a browser within three months of latest, are not operating any plugins known to sniff or record keystrokes and clicks, and are not running plugins with known vulnerabilities.

Unlike Zero Trust, the company does not reasonably own the client side of the conversation in a WEI context. This radically changes the power dynamics between public users and private servers. Under current internet rules, both sides mutually distrust each other and slowly establish trust. Under WEI, the server sets a minimum trust boundary that's far higher than is currently possible, which gives server operators far more power in this conversation than before. A Zero Trust like level of power, in fact.

What does Web-Environment-Integrity allow server operators to do?

As it happens, WEI is a clear example of a technique or standard that needs to have a clear and well thought out answer to the question:

What can a well resourced malicious actor do with this framework? How can they use this safety tool to harm people?

Right now, we don't have those answers. The explainer goes into some detail about how to avoid tracking through WEI techniques, but overlooks the thing that has everyone posting "Google wants to ban ad-blockers" headlines. The WEI proposal allows server operators to prohibit the use of specific browser plugins when accessing the site, which gives ad-networks an in-browser way to say "turn off your ad-blocker in order to see this content."

The well resourced malicious actor here is the internet advertising industry, of which Alphabet (Google's parent company) is one of the biggest members. The proposal writers do not view code injection and people-tracking through advertising to be malicious, they see it as a perfectly legitimate way to pay for services.

"But it's not the server operator doing the banning, it's the attestor; and the attestor has no idea what's on the site!"

The WEI standard involves three parties: The browser making the request, the server hosting the site, and the 'attestor' service the server relies on to judge the integrity of the browser making the request. The "Google wants to ban ad-blockers" headline happens when an advertising-driven site uses an attestor service that considers browsers with ad-block plugins to be insecure. Technically it isn't the server making the "no ad-block" constraint, at least at the time of the request. The Server operator made that choice when they selected an attestor service that prohibits ad-block plugins.

This sort of deniability is all over the tech industry.

I ran into a pretty common attitude regarding workplace diversity the other day. It was on a Q/A site. Paraphrased, the issue is:

Q: How can we improve the diversity of the candidates we hire?

A: That actually hurts the diversity of your hiring pool, because many people see "diversity!" on a hiring page and immediately go somewhere else. Who wants to be hired to a place that'll give a job to an unqualified minority just to meet some numbers?

The mechanism this answerer was assuming, that diversity programs are quota systems, has been explicitly illegal in the US since the 1980s. The Supreme Court at the time ruled that quotas like that were what this answerer said: racism, even if it was intended to correct for systemic biases. If you find your prospective workplace is using quotas, or explicitly hiring following racialized patterns, you have strong grounds for a lawsuit.

Within the last year, news broke of a company that got into hot water by someone posting "I'm hiring for X, give me all of your non-binary, queer, women, and minority leads please!" to Twitter. The strong implication by this statement was that this company was using a racialized hiring process which is illegal.

These days hiring pipelines at large US companies are engineered to avoid getting sued, and therefore don't use quotas. To build a hiring pipeline that furthers a company's diversity goals, while also avoid getting sued, requires several things:

  • You must interview/treat everyone who applies equally.
  • You must assess each application the same way.
  • You must make your final hire/pass decision based on the merits of the application.
  • (US) You must give preference to military veterans, by law.

So far, this is an "equity" argument. But building a system to improve your workplace diversity needs a few more steps.

  • You can change the mix of your applicants by biasing where you advertise the job.
  • You can't hide the job posting and pass out application links to your preferred groups. You still need to post them on your jobs page.
  • Remove biasing language from your posting and job application process.
  • You still have to treat all applicants the same once they've applied.

Equity, in other words.

Furthermore, some companies are beginning to reframe their diversity programs towards a "meet the market" approach. In that they assess their diversity program success based on how well their employee mix matches the potential job market for their roles. If a given position is 82% male in the job market, that's the target they'll push for; not 50%.

Equity, because that's the most legally conservative option if you want to avoid lawsuits for discrimination in US courts.

Mathew Duggan wrote a blog post on June 9th titled, "Monitoring is a pain" where he goes into delicious detail around where observability, monitoring, and telemetry goes wrong inside organizations. I have a vested interest in this, and still agree. Mathew captures a sentiment I didn't highlight enough in my book, that a good observability platform for engineering tends to get enmeshed ever deeper into the whole company's data engineering motion, even though that engineering observability platform isn't resourced enough to really serve that broader goal all that well. This is a subtle point, but absolutely critical for diagnosing criticism of observability platforms.

By not designing your observability platform from the beginning for eventual integration into the overall data motion, you get a lot of hard to reduce misalignment of function, not to mention poorly managed availability assumptions. Did your logging platform hiccup for 20 minutes, thus robbing the business metrics people of 20 minutes of account sign-up/shutdown/churn metrics? All data is noisy, but data engineering folk really like it if the noise is predictable and thus can be modeled out. Did the metrics system have an unexpected reboot, which prevented the daily code-push to production because Delivery Engineering couldn't check their canary deploy metrics? Guess your metrics system is now a production critical system instead of the debugging tool you thought it was.

Data engineering folk like their data to be SQL shaped for a lot of reasons, but few observability and telemetry systems have an SQL interface. Mathew proposed a useful way to provide that:

When you have a log that must be stored for compliance or legal reasons, don't stick it into the same system you use to store every 200 - OK line. Write it to a database (ideally) or an object store outside of the logging pipeline. I've used DynamoDB for this and had it work pretty well by sticking it in an SQS pipeline -> Lambda -> Dynamo. Then your internal application can query this and you don't need to worry about log expiration with DynamoDB TTL.

Dynamo is SQL-like, so this method could work for doing business metrics things like the backing numbers for computing churn rate and monthly active users (MAU) numbers. Or tracking things like password resets, email changes, and gift-card usage for your fraud/abuse department. Or all admin-portal activity for your annual external audits.

Mathew also unintentionally called me out.

Chances are this isn't someones full-time job in your org, they just happened to pick up logging. It's not supposed to be a full-time gig so I totally get it. They installed a few Helm charts, put it behind an OAuth proxy and basically hoped for the best. Instead they get a constant flood of complaints from consumers of the logging system. "Logs are missing, the search doesn't work, my parser doesn't return what I expect".

That's how I got into this. "Go upgrade the Elasticsearch cluster for our logging system" is what started me on the road that lead to the Software Telemetry book. It worked for me because I was at a smaller company. Also our data people stuck their straw directly into the main database, rather than our logging system, which is absolutely a dodged bullet.

Mathew also went into some details around scaling up a metrics system. I spent a chapter and part of an appendix on this very topic. Mathew gives you a solid concrete example of the problem from the point of view of microservices/kubernetes/CNCF architectures. This stuff is frikkin hard, and people want to add things like account tags and commit-hash tags so they can do tight correlation work inside the metrics system; the sort of cardinality problem that many metrics systems aren't designed to support.

All in all Mathew lines out several common deficiencies in how monitoring/observability/telemetry is approached in many companies, especially growing companies:

  • All too often, what systems are in place are there because someone wanted it there and made it. Which means it was built casually, and probably not fit for purpose after other folks realize how useful they are, usage scope creeps, and it becomes mission critical.
  • These systems aren't resourced (in time, money, and people) commensurate with their importance to the organization, suggesting there are some serious misalignments in the org somewhere.
  • "Just pay a SaaS provider" works to a point, but eventually the bill becomes a major point of contention forcing compromise.
  • Getting too good at centralized logging means committing to a resource intensive telemetry style, a habit that's hard to break as you get bigger.
  • No one gets good at tracing if they're using Jaeger. Those that do, are an exception. Just SaaS it, sample the heck out of it, and prepare for annual complaining when the bill comes due.

The last point about Jaeger is a key one, though. Organizations green-fielding a new product often go with tracing only as their telemetry style, since it gives so much high quality data. At the same time, tracing is the single most expensive telemetry style. Unlike centralized logging, which has a high quality self-hosted systems in the form of ELK and Loki, tracing these days only has Jaeger. There is a whole SaaS industry who sell their product based on how much better they are than Jaeger.

There is a lot in the article I didn't cover, go read it.

The trajectory of "AI" features

I've been working for bay area style SaaS providers for a bit over a decade now, which means I've seen some product cycles come and go. Something struck me today that I want to share, and it relates to watching product decisions get made and adjusting to market pressures.

All the AI-branded features being pushed out by everyone that isn't ChatGPT, Google, or Microsoft all depend on one of the models from those three companies. That, or they're rebranding existing machine learning features as "AI" to catch the marketing moment. Even so, all the features getting pushed out come from the same base capabilities of ChatGPT style large language models:

  • Give it a prompt, it generates text.

Okay, that's the only capability. But this one capability is driving things like:

  • Rewriting your thing to be grammatical for you!
  • Rewriting your thing to be more concise!
  • Suggesting paragraphs for your monthly newsletter!
  • Answering general knowledge questions!

That's about it. We've had about half a year of time to react to ChatGPT and start generating products based on that foundation, and the above is what we're seeing. A lot of these are one to three engineering teams worth of effort over the course of a few months, with a month or three of internal and external testing. These are the basic features of LLM-based machine learning, also known right now as AI.

We don't yet know what features we'll have in two years. "Suggest text from a prompt" went from hot new toy to absolute commodity table-stakes in 8 months, I've never seen it before. Maybe we've fully explored what these models are capable of, but we sure haven't explored the regulatory impacts yet.

The regulatory impacts are the largest risk to these tools. We've already seen news stories of lawyers using these tools to write briefs with fictional citations, and efforts to create deep fake stories "by" prolific authors. The European Union is already considering regulation to require "cite your sources", which GPT isn't great at. Neither are GPT-based models good about the Right To Be Forgotten enshrined in EU law by GDPR.

The true innovation we'll see in coming years will be in creating models that can comply with source-citing and enable targeted excludes. That's going to take years, and training those models is going to keep GPU makers in profit margins.

Jeff Martins at New Stack had a beautiful take-down of the SaaS provider practice of not sharing internal status, and how that affects down-stream reliabiilty programs. Jeff is absolutely right, each SaaS provider (or IaaS provider) you put in your critical path decreases the absolute maximum availability your system can provide. This also isn't helped by SaaS providers using manual processes to update status pages. We would all provide better services to customers if we shared status with each other in a timely way.

Spot on. I've felt this frustration too. I've been in the after-action review following an AWS incident when an engineering manager asked the question:

Can we set up alerting for when AWS updates their status page?

As a way to improve our speed of response to AWS issues. We had to politely let that manager down by saying we'd know there was a problem before AWS updates their page, which is entirely true. Status pages shouldn't be used for real-time alerting. This lesson was further hammered home after we gained access to AWS Technical Account Managers, and started getting the status page updates delivered 10 or so minutes early, directly from the TAMs themselves. Status Pages are corporate communications, not technical.

That's where the problem is, status pages are corporate communication. In a system where admitting fault opens you up to expensive legal actions, corporate communication programs will optimize for not admitting fault. For Status Pages, it means they only get an outage notice after it is unavoidably obvious that an outage is already happening. Status Pages are strictly reactive. For a true real time alerting system, you need to tolerate the occasional false positive; for a corporate communication platform designed to admit fault, false positives must be avoided at all costs.

How do we fix this? How do we, a SaaS provider, enable our downstream reliability programs to get the signal they need to react to our state?

There aren't any good ways, only hard or dodgy ones.

The first and best way is to do away with US style adversarial capitalism, which reduces the risks of saying "we're currently fucking up." The money behind the tech industry won't let this happen, though.

The next best way is to provide an alert stream to customers, so long as they are willing to sign some form of "If we give this to you, then you agree to not sue us for what it tells you" amendment to the usage contract. Even that is risky, because some rights can't be waved that way.

What's left is what AWS did for us, have our Technical Acount Managers or Customer Success Managers hand notify customers that there is a problem. This takes the fault admission out of the public space, and limits it to customers who are giving us enough money to have a TAM/CSM. This is the opposite of real time, but at least you get notices faster? It's still not equivalent to instrumenting your own infrastructure for alerts, and is mostly useful for writing the after-action report.

Yeah, the SaaSification of systems is introducing certain communitarian drives; but the VC-backed capitalistic system we're in prevents us from really doing this well.

For many reasons, discussions at work have turned to recovering/handling layoffs. A coworker asked a good question:

Have you ever seen a company culture recover after things went bad?

I have to say no, I haven't. To unpack this a bit, what my coworker was referring to is something loooong time readers of my blog have seen me talk about before, budget problems. What happens to a culture when the money isn't there, and then management starts cutting jobs?

In the for-profit sector of American companies, you get a heck of a lot of trauma from the sudden-death style layoffs (or to be technical reductions in force because there is no call-back mechanism in the SaaS industry). Sudden loss of coworkers with zero notice, scrambling to cover for their work, scrambling to deal with the flash reorganization that almost always comes with a layoff, scrambling to worry about if you'll be next, it all adds up to a lot of insecurity in the workplace. Or a lack of psychological safety, and I've talked about what happens then.

This coworker was also asking about some of the other side effects of a chronic budget crunch. For publicly traded companies, you can enter this budget crunch well before actual cashflow is a problem due to the forcing effects of the stock market. Big Tech is doing a lot of layoffs right now, but nearly every one of the majors is still profitable, just not profitable enough for the stock market. Living inside one of these companies means dealing with layoffs, but also things like:

  • Travel budget getting hard to approve, as they crank down spend.
  • Back-filling departures takes much longer, if you get it at all.
  • Not getting approval to hire people to fill needs.
  • Innovation pressure: have to hit bolder targets with the same resources, and the same overhead.
  • Adding a new SaaS vendor becomes nearly impossible.
  • Changing the performance review framework to emphasize "business value" over "growth".
  • Reduced support for employee groups like those supporting disabled and racial employees.
  • Reorgs. So many reorgs.
  • Hard turns on long term strategy.

If a company hasn't had to live through this before the culture doesn't yet bear the scars. The employees certainly do, especially those of us who were in the market in 2007-2011. Many of us left companies to get away from this kind of thing. That said, the last major down period in tech was over ten years ago; there are a lot of people in the market right now who've never seen it get actually bad for a long period of time. "Crypto Winter", the fall of many crypto-currency based companies, was as close as we've gotten as an industry.

When the above trail of suck starts happening in a company that hasn't had to live through it before, it leaves scar tissue. The scars are represented by the old-timers, who remember what upper management is like when the financial headwinds are blowing strong enough. The old salts never regain their trust of upper management, because they know first-hand that they're face eating leopards.

Even if upper management turns the milk and honey taps on, brings out the bread and circuses, and undoes all the list-of-suck from above, the old-timers (which means the people in middle management or senior IC roles) will still remember upper management is capable of much worse. That changes a culture. As I talked about before, trust is never full again, it's always conditional.

So no, it'll never go back to the way it was before. New people may think so, but the old timers will know otherwise.

...and why this is different than blockchain/cryptocurrency/web3.

Unlike the earlier crazes, AI is obviously useful to the layperson. ChatGPT finished what tools like Midjourney started, and made the average person in front of a browser go, "oh, I get it now." That is something Blockchain, Crypto currencies, and Web3 never managed. The older fads were cool to tech nerds and finance people, but not the average 20 year old trying to make ends meet through three gig-economy jobs (except as a get-rich-quick scheme).

Disclaimer: This post is all about the emotional journey of AI-tech, and isn't diving into the ethics. We are in late stage capitalism, ethics is imposed on a technology well after it has been on the market. For a more technical take-down of generative AI, read my post from April titled "Cognitive biases and LLM/AI". ChatGPT-like technologies are exploiting human cognitive biases baked into our very genome.

For those who have avoided it, the art of marketing is all about emotional manipulation. What emotions do your brand colors evoke? What keywords inspire feelings of trust and confidence? The answers to these questions are why every 'security' page on a SaaS product's site has the phrase "bank-like security" on it; because banks evoke feelings of safe stewardship and security. This is relevant to the AI gold rush because before Midjourney and ChatGPT, AI was perceived as "fancy recommendation algorithms" such as those found on Amazon and the old Twitter "for you" timeline; after Midjourney and ChatGPT AI became "the thing that can turn my broken English into fluent English" and was much more interesting.

The perception change caused by Midjourney and ChatGPT is why you see every tech company everywhere trying to slather AI on their offerings. People see AI as useful now, and all these tech companies want to be seen as selling the best useful on the market. If you don't have AI, you're not useful, and companies who are not useful won't grow, and tech companies that aren't growing are bad tech companies. QED, late stage capitalism strikes again.

It's just a fad

Probably not. This phase of the hype cycle is a fad, but we've reached the point where if you have a content database 10% the size of the internet you can algorithmically generate human-seeming text (or audio, or video) without paying a human to do it; this isn't going to change when the hype fades, the tech is here already and will continue to improve so long as it isn't regulated into the grave. This tech is an existential threat to the content-creation business, which includes such fun people as:

  • People who write news articles
  • People who write editorials
  • People who write fiction
  • People who answer questions for others on the internet
  • People who write HOW TO articles
  • People who write blog posts (hello there)
  • People who do voice-over work
  • People who create bed-track music for podcasts
  • People who create image libraries (think Getty Images)
  • People who create cover art for books
  • People who create fan art for commission

The list goes on. The impact here will be similar to how streaming services affected musician and actor income streams: profound.

AI is going to fundamentally change the game for a number of industries. It may be a fad, but for people working in the affected industries this fad is changing the nature of work. I still say AI itself isn't the fad, the fad is all the starry-eyed possibilities people dream of using AI for.

It's a bullshit generator, it's not real

Doesn't matter. AI is right often enough to fit squarely into human cognitive biases of trustworthy. Not all engines are the same, Google Bard and Microsoft Bing have some famous failures here, but this will change over the next two years. AI answers are right often enough, and helpful often enough, that such answers are worth looking into. Again, I refer you to my post from April titled "Cognitive biases and LLM/AI".

Today (May 1, 2023) ChatGPT is the Apple iPhone to Microsoft and Google's feature-phones. Everyone knows what happened when Apple created the smartphone market, and the money doesn't want to be on the not Apple side of that event. You're going to see extreme innovation in this space to try and knock ChatGPT off its perch (first mover is not a guarantee to be the best mover) and the success metric is going to be "doesn't smell like bullshit."

Note: "Doesn't smell like bullshit," not, "is not bullshit". Key, key difference.

Generative AI is based on theft

This sentiment is based on the training sets used for these learning models, and also on a liberal interpretation of copyright fair use. Content creators are beginning to create content under new licenses that specifically exclude use in training-sets. To my knowledge, these licenses have yet to be tested in court.

That said, this complaint about theft is the biggest threat to the AI gold rush. People don't like thieves, and if AI gets a consensus definition of thievery, trust will drop. Companies following an AI at all costs playbook to try and not get left behind will have to pay close attention to user perceptions of thievery. Companies with vast troves of user-generated data that already have a reputation for remixing, such as Facebook and Google, will have an easier time of this because users already expect such behavior from them (even if they disapprove of it). Companies that have high trust for being safe guardians of user created data will have a much harder time unless they're clear from the outset about the role of user created data in training models.

The perception of thievery is the thing most likely to halt the fad-period of AI, not being a bullshit generator.

Any company that ships AI features is losing my business

The fad phase of AI means just about everyone will be doing it, so you're going to have some hard choices to make. The people who can stick to this are the kind of people that are already self-hosting a bunch of things, and are fine with adding a few more. For the rest of us we have harm reduction techniques like using zero-knowledge encryption for whatever service we use for file-sync and email. That said, even the hold-out companies may reach for AI if it looks to have real legs in the marketplace.

Yeah. Like it or not, AI development is going to dominate the next few years of big-tech innovation.

I wrote this because I keep having this conversation with people, and this makes a handy place to point folk at.

Cognitive biases and LLM/AI

I've been a moderator on ServerFault (a Q&A site like StackOverflow) for over a decade at this point. I am deeply, intimately familiar with what people in a hurry to solve a problem look for. They want:

  • A how-to explaining exactly how to get from where they are to where they need to be.
  • No faffing about with judgment calls regarding the wisdom of doing so.

This is good for Q&A sites, because not everyone is good at generalizing their solution from 30 other answers similar but not quite like their case. They want a step by step how-to that will get them out of the hole they're in. The most trusted ServerFault answerers (I was among them once) were good at doing this. We didn't throw RTFM at them, or cite section and sub-section of the manual they should have read. We explained why it went bad, how to get out of the hole, and perhaps some next steps. The better ones weren't judgmental in describing the foot-guns in play, but described negative consequences that you might not want to experience.

ChatGPT and related technologies act like higher quality answerers. You ask it "give me a step by step guide to get to where I need to go," and that is what it will produce. A step by step guide, no judgement about any foot-guns you're playing with. So nice. You don't have to put up with the judgmental best practices chorus to do so, and it isn't judgmental every time. You don't have to hope to get a nice answerer the way you do on ServerFault/StackOverflow.

Whether that solution is actually fit for purpose is not guaranteed, but it sure looks authoritative. This is where the cognitive biases come in to play.

Humans looking for help want one thing: someone who knows how to do the thing to help them do the thing. We asses "knows how to do the thing" many ways, but a time tested strategy is buried in the definition of the word mansplaining. To save time, these are the qualities we look for as a species:

  • Is reassuring that this will actually work.
  • Is confident in their knowledge.
  • Is willing to go deeper if asked.

These three are extremely possible to fake, and humans have been doing so since humans were invented. The check and balance that Q&A sites like ServerFault and StackOverflow offer is voting on the answers, so actually wrong answers get some signal to validity. It have seen people be confidently wrong time and time again on ServerFault, and get cranky when their lack of actual knowledge gets called out.

Technology like ChatGPT automates producing answers that hit all three points, but lacks the voting aspect and any guarantee of actual expertise. In short, ChatGPT and related technologies automate the exploitation of human "sounds like an expert" heuristics. This is an exploit of the human cognitive system, and this is why you're going to start seeing it everywhere. It have complete faith that some startup somewhere is going to come up with the idea of:

What if we have generative LLM/AI create answers to questions that are then voted on by 'experts' and we sell the answers?

It's coming. May already be here.