- Microsoft will continue to grow. Their tech ecosystem is hyper-expansive, touching everything related to development and IT (.Net, TypeScript, Github, VSCode, PowerPlatform, Azure, etc.). Please don't directly come at me with moral qualms or commentary on the quality of their products - I'm just the messenger.
- Low-code tooling will see a growth spurt, not driven by "citizen developers" but by increased developer adoption of technologies that offer a good DX and a capability to integrate seamlessly into "normal" dev workflows. Tech like Retool, Plasmic, Builder.io, PowerPlatform, new-gen ETLs, etc.
- The "citizen developer" paradigm will remain largely unsolved. The vision holds all its initial potential but it requires a multi-pronged, universal, push involving corporate politics, compliance, security, training, etc. that cannot be solved by a single company. We might see interesting initiatives or companies built around this in 2023, but a year won't be enough time to solve it.
- There will be a massive proliferation of GPT-based apps. Many will be shite, a few will be useful.
- Rising interest rates will push investment mass away from startups and into the other end of the spectrum: good old bonds and other fixed-rate assets.
- There will be a rising movement of software dev methodologies and training on how to use AI-assisted coding (Copilot, ChatGPT) effectively.
- ChatGPT will not replace coders at least in 2023, and the bitter non-tech people who are expressing uninformed opinions on its code-gen capabilities, observing from the sidelines and waiting for some sort of comeuppance, are not going to get it.
- React will continue to be a standard but coding a form will be as difficult as it was back in fucking 2013. I love the tech otherwise, but c'mon.
yes. Everybody tends to think their business requirements are very unique. In reality not so much.
Also it does not need to replace -all- coders. but if ONE smart Architect is able to bang out a mid tier MVP all by himself by feeding a textbox, you might feel trouble approaching.
> Everybody tends to think their business requirements are very unique
How did that pan out for low-code tools? Or hell, the advent of high-level languages in the 60's and 70's? Did either of these paradigm shifts reduce the amount of coders needed?
For these tools to really eat up into demand for programmers, it would need to increase productivity enough to:
1. Meet the ever-growing demand for software
2. Close the permanent gap for devs in the world (which some quantify in the millions)
Dev prospects could worsen in 2023 due to the rise of interest rates and the displacement of investment towards fixed-assets, but if you think that Davinci or whatever model is the factor that's going to be pushing the needle, I don't know what to tell you.
The growth won't last forever. At least not at this pace.
How many companies write the same bits of software as everyone else, destined to throw them away after a year or two? How much of that waste is driven by continuous churn of frameworks? How much of that software itself is pointless, powering businesses that should never be?
I'm thinking of OP's prediction:
> - Rising interest rates will push investment mass away from startups and into the other end of the spectrum: good old bonds and other fixed-rate assets.
If that were to happen, I wonder what it'll do to the "ever-growing demand for software".
The other thing people don't realize is if one "architect" is able to bangout an mvp super fast with an AI what is stopping others? Then wouldn't the over supply just mean mvps are now severely commoditized? Youve just moved the demand else where. Instead of "idea people" we will now have "mvp people"?
you mean the low code tools like wix that make billions?
Look I see what you mean, but even you have to admit that helping generate the basic structure for something will save days of development and the fact that it can probably learn a new framework of tool xyz quicker than us will give a big boost.
I'll make an anti-prediction instead. People will fail to find decent use cases for ChatGPT/etc. We'll get used to the fact that AI models can generate language that looks indistinguishable from humans on the surface, but there won't be profound impact. GPT-3 was released in 2020, and generative image models have been decent since ~2017. In the short term there hasn't been as huge of an impact as many have predicted. It'll probably happen very slowly over the next 5-10 years. People overestimate the short term.
Twitter is going to be interesting to watch. I think it'll either crash and burn, or become huge doing something very different from what it is now.
EDIT: I'll add one more. Meta will keep declining and fails to deliver anything meaningful wrt. Metaverse.
I agree that people will get used to text generation existing and it won't be as exciting. But I think people are already finding lots of use cases for it.
For example, we had a case where we had thousands of technical medical descriptions and we wanted to simplify them into "average person" descriptions. In the past, that meant someone going through each one and re-wording them as best as they could. But an engineer tried pasting them all in to ChatGPT and asking it to reword them and it worked surprisingly well.
In another example, my wife had to send an email to a volunteer group asking them to participate in an upcoming activity. She was able to draft a base email to edit in about 1 second with ChatGPT. Then she could edit it as needed.
I've also heard from folks in marketing getting a lot of use out of these tools.
So maybe none of these use cases are individually transformative, but I think it will at least be a tool on par with spellcheck that is super handy and widely used.
I agree. What I meant to say is not that there won't be any use cases, but rather that they won't be as obviously transformative as people expect them to be. We won't have new kinds of products that are uniquely enabled by ChatGPT. Instead, we will get incremental improvements in spell checking, rewording text, code search, etc. But just like Gmail's autocomplete, better language translation, or better voice recognition, these will be largely behind the scenes and won't lead to big societal changes. Many of these improvements will be existing companies integrating better models, not new startups that become possible now because of ChatGPT.
I’m not an expert in anything, just wild guesses projecting some of my biases:
1- significant portion of mass market retail becomes automated. Labor is getting too expensive and entitled. Everyone needs to hire no one wants to work and businesses can’t make $20 minimum wages profitable.
2- Drone military spending goes up by 10-100x. Fueled again by enlistment shortages and arms proliferation and need for deterrence.
3- AR/VR becomes the most hyped consumer technology (more than now). People are thirsting for something new and a 100 megapixel camera in the phone doesn’t move the needle
4- More and more companies announce migration away from Chinese manufacturing. Vietnam and India become primary beneficiaries. Global undeclared economic war wages on.
5- Google and Fb will see persistent losses in active usage. They are too addicted to the ad money and user experience is becoming borderline hostile.
6- Still won’t be able to buy PS5 if you want one. Nintendo will continue to milk the switch without a meaningful update. Won’t stop unless sales dry up.
General Trend: Balkanization of the Internet continues and significantly accelerates.
Specific Prediction: Google services are banned by a number of countries, and separately their search business is disrupted by AI generated content to the point that their algorithms stop working as intended.
It's not such an outrageous guess, Russia has been building up the army presence at the Ukrainian border for months before the invasion, citing military exercises as the official reason.
You mean tokens based on blockchains? They have one (and only one) real and worldwide successful use case - law avoidance. Any law, in any country, by any participant. The moral value of this system is an open question though.
We are going to have massive shift towards rejuvenation. That is the thing that everybody wants including countries as a whole.
Who wouldn't want to reverse aging and almost fix most of the disease and costs associated with it on their population. Who, on individual level, wouldn't want to look 20 again.
Right now, the problem seems to be that we, in general, we do not believe that such thing is possible.
This will change as soon as we have a robust rejuvenation event at least in mice. After that, the rejuvenation is going to be the new field where all the
investors money will go.
On a more modest perspective, stuff related to health and longevity will also see growth. New sensors, new ways to gradually improve health using tech.
I would say, that it is less of a prediction for 2023 and more of a trend. Like. I think about the next decade being predominately health/longevity focused.
There is also some interesting application for statistical models like ChatGPT and beyond. I would say, that text generation is not the most interesting application for such models.
For example, what kind of properties we could infer from our genome using network models.
- Big Tech becoming more privacy-friendly (just like Apple who just enabled end-to-end encryption on iCloud)
- Mostly a lot of the same stagnation we've seen in the tech industry for the past 5 years.
- Some sprouts of innovation around generative language, probably around fact checking, detecting and removing "hallucinations" from models. But probably not anything significant enough to significantly shift anything for every day users (probably have to wait 2025 for that).
- Crypto folks will keep believing that they're building the future, although everything indicates that Big Tech is finding a way to become safer, more private, without having to become fully decentralized.
As I shared, Apple just enabled E2E on iCloud. Dropbox also just acquired BoxCryptor. People increasingly demand privacy and use third party services that provide it. It's actually not that hard or expensive to implement. It seems to me that Big Tech is realizing that, and not wanting to lose market share, thus moving in that direction. There are also many improvements in AI differential learning and other techniques that can be used to still learn from data and exploit value from it, while respecting people's privacy (Example: your iPhone recognizes objects and people in pictures, but the data is never sent to Apple.)
Regarding code generation from design, this space is more or less as old as designing webpages itself. Even early versions of Adobe Photoshop already included the slice functionality which you could then export as HTML. This part has always been reasonably easy. The real problem is how do you keep your changing design in sync with your changing code and I'm yet to see anyone solve that problem well.
Another problem with code generation is that maybe you want React code, I want static html, someone else wants Vue.js code, etc. There isn't a single standard target that everyone agrees on and the frontend landscape change so frequently that it would be hard to keep up.
A lot of times you also have a page template framework in your app with a reusable header, footer, etc, and are really just designing the content of a page. It's difficult for generated code to slip cleanly in to an existing page framework in a way that doesn't take more time to rework than to just write yourself.
I'm not saying these are unsolvable problems, but it just makes it hard for a code generation product like this to find a single big enough addressable market to be exciting.
Java (both the language, but especially the platform) will become even more popular. The work done on it is phenomenal, it’s getting better FFI, Loom is a game changer and value types may actually start to make a debut next year. All this with a blooming language ecosystem, and I didn’t even mention GraalVM, which allows even more language to target the platform, sometimes with better speed then they would otherwise have, complete with polyglot interop.
I think everyone focuses on the major trends. But there are important background/side trends.
As people realize the drawbacks of silos and the benefit of more intentionally crafted data, Linked Data and other network relative data schemes will become more popular, in addition to their uses for AI systems (which should be the biggest "site" of them all, and something LD is created for).
I would characterize AI as 80% accurate so far. But getting one in five things wrong is not good enough for many tasks. Human/machine oriented data formats like Linked Data will help close this gap, as contributed by projects like Wikidata and increasingly smaller scale apps through better defined SEO (schema.org), for example. Breakthroughs in easily working with Linked Data at day to day levels would be helpful here, right now libraries even for specific domains are very nuts and bolts compared to ORM libraries. For common querying, perhaps GraphQL with network schemas will start to gain mainstream popularity.
We should also see breakthroughs in open standards data carrier formats, like decentralized wallets and credentials. These will have have significant impact because they are essentially like free-floating sites that interact with any site.
Okay, you said “apart from ChatGPT”, but that doesn’t seem fair and since nobody said it yet:
I think 2023 is the year AGI becomes a serious and mainstream topic. There will be new versions of OpenAI’s work, that will incorporate layers for correctness checking and reinforcement learning. Versions of itself that start to improve off each other.
As technical limitation after technical limitation is solved or lifted, HN and engineers at large start having discussions about what it means to be an AGI in no true Scotsman style (“no, if it can’t exactly be taught how to create a startup, fundraise for it and have a 7bn dollar exit then it’s not REAL AGI”), until at some point in the next years we stop caring about that discussion as we moved on from “are we there yet” to “what now”.
This will be longer than a year as an ordeal. I predict we will look back at 2023 as when it all started. Even though, as we all know, it didn’t, GPT had been around for a while and has been building on the shoulders of giants; but for the mainstream, it absolutely hasn’t started yet.
- WFH will increase and thus apps and processes that support this. For example, I expect we have one additional team communication product on the Slack/Teams level of popularity. Focusing on either privacy or more employer surveillance.
- Legal issues for AI coding help.
- Commercial support/add-ons for the "fediverse". Resulting in a lot of de-federalization and thus at least two sub-spheres.
- AAA gaming will tackle the new rise of AI. Better bots, for one (your computer will now insult your mom, too). Prompting tech journalists to coin horrible new pseudo-acronyms like "AIAIAI" or "AAAAI".
- More people moving away from VSC due to better LSP integration in existing and new editors (Helix, neovim etc.)
- You'll read a lot more about "permacomputing", with no definite products.
- Metaverse will continue to fail, possible new book by Jared Lanier gloating.
I don't expect ChatGPT will have much impact in 2023 or in general. ChatGPT etc, will be used by some devs as just as some devs like IDEs vs editors, or how often devs search the web or SO. I've rarely ever been limited by how long it takes to implement an idea, or coming up with an idea when I could clearly express the problem.
Edit: What would be cool is if ChatGPT et al, could take an implementation as input and generate all the test descriptions in natural language (source too, why not). That would expose unintended hidden or mistaken assumptions in the implementation.
in an age of computer generated content, public websites will become garbage heaps of object spam no matter how much verification and moderation you throw at it
the only interesting spaces will be anonymous and invite-only
the internet becomes balkanized as international cyberattacks increase in frequency
I’d love to see one of the FAANGs endorsing and/or promoting or liberating their tech docs solutions, including better lightweight markup for docs (not that Asciidoc and rST are bad, but they aren’t super popular yet). JetBrains is about to do that, for example: https://lp.jetbrains.com/writerside/
Nothing particularly interesting outside of AI. But, a lot of interesting new products enabled by the integration of GPT-3, maybe even in the crypto scene.
For example, there have always been ideas floating around about creating a blockchain based repository of scientific knowledge or social media or whatever. The main blockers for ideas like that were always content moderation at scale: how do you know that what someone is uploading is legitimate or follows the rules? How do you moderate content without an overlord? There need to be systems in place to:
- filter out garbage and toxicity,
- allow heterodox submissions (no political censorship),
- and do it all without some kind of admin/moderator
Now GPT-3 can be used to pre-validate content. It can tell you whether a submission follows some kind of logical reasoning, isn't spam, and isn't anti-social. It's not 100% accurate, but it's good enough, and it can be tweaked.
Beyond that, GPT-3 can be used for all sorts of tools like automated codebase documentation (no more writing for developers, yay!), news trend-spotting (for traders/finance), generating landing page copy, etc.
I find these kinds of descriptions confusing. What does "garbage and toxicity" mean? You might as well have written "filter out bad things", it would be equally vague and subjective.
Not exactly. It's possible to identify anti-social comments. You just have to define what's considered anti-social, either from a human nature perspective or a cultural one.
For example, if we go the human nature route, then any comment - no matter how inflammatory - is fair game as long as the goal of the comment is to "help the tribe" in some sense. That's kind of like political speech. So, you could faithfully argue ideas like "nazis are good, and we should be like them", as long as you aren't using toxic language and attacks on other people while doing it.
EDIT: Actually fascist/communist ideas contain beliefs that are anti-social, so wouldn't be considered pro-social according to human nature, and that's where the conversation would stop. It still stands that heterodox ideas that don't work, or people don't like, can be technically pro-social.
If we go the cultural route, then identify the mainstream beliefs and rules of discussion and enforce them. This is like the "no swearing" rule, or "no nazis" rule.
You could just codify a set of rules for content that the AI adheres to.
By human nature, anti-social would be anything that's meant to be destructive to, or an attack on "the tribe" or a member of the tribe. So, some examples:
- "Orange people are useless."
- "Fuck you, you piece of shit."
- "<insert political party> voters are idiots."
- "We should take away the rights of <insert demographic>."
The more sophisticated the comment the more difficult it is to tell whether it's overall anti-social or not, but we can at least identify parts of a comment that are anti-social and flag them as such.
The main issue would be people trying to get around the AI's ability to recognize it by obfuscation.
Let's take Twitter for example. There's new information coming out that Twitter colluded with the US government to shadow-ban and censor political opponents (including politicians) in the last 4-5 years, without anyone knowing about it, or being able to prove it.
That wouldn't be possible on a blockchain-based social platform because the algorithm would be public, provably in use, and there would likely be voting mechanisms for changes to the algorithm.
- Low-code tooling will see a growth spurt, not driven by "citizen developers" but by increased developer adoption of technologies that offer a good DX and a capability to integrate seamlessly into "normal" dev workflows. Tech like Retool, Plasmic, Builder.io, PowerPlatform, new-gen ETLs, etc.
- The "citizen developer" paradigm will remain largely unsolved. The vision holds all its initial potential but it requires a multi-pronged, universal, push involving corporate politics, compliance, security, training, etc. that cannot be solved by a single company. We might see interesting initiatives or companies built around this in 2023, but a year won't be enough time to solve it.
- There will be a massive proliferation of GPT-based apps. Many will be shite, a few will be useful.
- Rising interest rates will push investment mass away from startups and into the other end of the spectrum: good old bonds and other fixed-rate assets.
- There will be a rising movement of software dev methodologies and training on how to use AI-assisted coding (Copilot, ChatGPT) effectively.
- ChatGPT will not replace coders at least in 2023, and the bitter non-tech people who are expressing uninformed opinions on its code-gen capabilities, observing from the sidelines and waiting for some sort of comeuppance, are not going to get it.
- React will continue to be a standard but coding a form will be as difficult as it was back in fucking 2013. I love the tech otherwise, but c'mon.