2026 Predictions

[dire thistle]

g

[tired gust]

World models will be a big thing in 2026

One major AI company will get acquired

Also im 3/3 im the best predictor @dire thistle

[dire thistle]

• AI psychosis is going to get much worse, will be treated as a public health crisis
• local LLMs will become more accessible. the bar will be so low that normies can hop in with 0 technical knowledge
• local LLMs becoming more accessible will also make AI psychosis worse

[dire thistle]

sure, but 7/8 is more correct predictions

a buncha ppl made 1/1s

[simple glen]

i have no idea what's gonna happen at this point everything is so unpredictable

[dire thistle]

• renewing my previous year's prediction: I think an act of violence against a prominent figure in AI will happen. yeah I was wrong last year but I won't be wrong this year (trust me bro)
• the white collar job market being kinda shit in 2025 will be conclusively linked to AI automating people out of work
• white collar job market will be even worse in 2026
• at least one white collar job that we know now will become more or less completely automated. it will no longer exist whatsoever. TOTAL JOB DEATH
• UBI will start to be treated as a more reasonable proposition, but at least in the US, we will make ZERO progress towards implementing it as policy

• videoslop will get much worse. it will become endemic on every platform. technically inclined users will still be able to recognize it, but only barely; normies will completely miss that iit's AI

@queen brook hop in

[dire thistle]

well predict anyway. it's free 😎

[simple glen]

what if we make it not tho 🤫

let's make a polymarket for each prediction

[dire thistle]

• AI detection tech like Pangram will initially grow and become even better and more reliable than they are now! but later in the year, it'll be unceremoniously killed, as RL training pipelines become easier to use, and commercial interest in "undetectable" AI grows

[dire thistle]

• kinda unrelated to AI - prediction markets will explode in popularity, becoming the dominant way gamblers blow their life savings

@mellow hearth @scarlet tinsel get in here too

[mellow hearth]

• there will be a major scandal in one or more national governments involving AI hallucinations messing up public services

[dire thistle]

ooh good one

[wintry creek]

Gpt-6 will not be released.

Steam will be full of AI generated games.

[mellow hearth]

• there will be a mass shooting incident at an AI datacentre or company office

[dire thistle]

@queen brook will be forced to change his username to GPT-7 after the real GPT-6 releases, because people will ping him in here thinking he's a bot all the time

[wintry creek]

A middle eastern country bans sex robots for becoming closer to mainstream

[simple glen]

1. we will not get a new breakthrogh technology at about the level of reasoning (that's generally used in models)
2. there will be a new useless use case of ai like computer use
3. claude code will continue being sota for agenetic coding
4. we will get a gpt 4.5 style model from openai
5. some (usable) open source model will surpass all closed models on the artificial analysis index

[dire thistle]

• computer use will stop being useless garbage and start being actually relevant in day to day usage
• it'll probably be through something like a browser extension, as opposed to a feature on a website like openai's agent mode currently is

[simple glen]

6. there will still not be any good implementation of native live video

[simple glen]

7. computer use won't stop being useless garbage and will die out

[dire thistle]

@brave star get in here and predict

[wintry creek]

AI animated anime will be quite common.

[simple glen]

8. half videoslop will be really bad made and therefore considered slop but the other half will be so unrecognizable it will mix with other content

[dire thistle]

• a big youtuber will have a "scandal" where anti-AI fans discover that parts of their videos (which they otherwise really enjoyed) were AI-generated

[simple glen]

9. there will be no real use case for ai images and ai videos outside of very small things like replacing stock images

[simple glen]

you are point farming that's just gonna happen 😡

[dire thistle]

ooh good one. i feel like we're still a ways off but this is realistic

[wintry creek]

The og digital anime was really shit, also one frame man convinces me it will be a thing.

ai can do better than one frame man

[brave star]

Reasoning models will evolve into either not human readable or barely readable 'efficient' reasoning.

[simple glen]

10. openai will become less of a bitch with access to their models and summaries

11. a new open source lab becomes prominent (stricken)

[simple glen]

scratch that that's too simple

[dire thistle]

I decided to show Gemini 3 Flash the predictions + outcomes from last year and ask it for its predictions this year. not gonna copy paste its huge wall of text but here are its predictions:

• RL will become extremely dominant because we'll be completely out of uncollected data from the human internet
• humanoid "model T" type robot will see the first major wave of deployments in real work environments
• AI wearables will catch on in a big way
• senior level positions will be fine in white collar work but juniors will face major job market issues
• at least 1 full length feature film will be made entirely with AI and it won't be a tech demo, it'll actually be good
• search engine use will fall off a cliff
• first solo billionaire will come from deployment of AI agents
• digital ID will be deployed for "proof of humanity"
• prices will drop by 99% for equivalent levels of today-intelligence

[simple glen]

11. at least two new open source labs become prominent

[dire thistle]

I'm gonna go the complete opposite direction on this. OpenAI will become even more trigger happy with bans for people who try to leak internal COT and will require even more stringent ID verifications to use their new models

[simple glen]

12. there will be a anti ai feature in some social media that blocks all ai content

[dire thistle]

god i hope ur right on this one

[simple glen]

tent is anti ai

13. we will achieve singularity and asi and the Internet will fall because of a rogue asi and humans will create a counter ai called the blackwall which will live distributed on every device protecting it from rogue ais

14. there will be the first malicious hack done by a agent that will steal hundreds of thousands of dollars

15. we will get a first non amateur ai image detector (not relying on watermarks)

16. at least 30 new anti ai lawsuits

17. openais new model is gonna be called o4 or o5 or o6 (stricken)

18. openai is gonna do a new weird change to the api semantics like they changed "minimal" to "none"

19. groq releases a new generation of their chip

20. antigravity will not take off and will die

21. google will not catch up on tool use to openai and anthropic

22. meta does not release anything at all

23. non ai: someone will claim the rapture again and it will be a big thing

[queen brook]

most of the predictions youve posted so far are things which will halfway happen but not fully happen

• everyone will agree that models by the end of 2026 will be agi for digital/computer tasks

[simple glen]

24. everyone will think that the current models are not agi but agi is just behind the door

[dire thistle]

@queen brook is right i need to get bolder. sora app will shut down

[dire thistle]

o-series naming convention will go away. no more o-models

[queen brook]

no i meant the opposite

you should be less bold

[simple glen]

25. openai will not make a non reasoning model and focus on hybrids

[simple glen]

honestly i agree im striking that

[simple glen]

26. i will keep my daily com streak until the end of the year

27. we get a new hype modality that everyone uses for 2 weeks and then it dies out (like image edit or videogen)

28. microsoft releases MAI to the general public

29. nvidia makes a hyper llm optimized chip like groq

30. jup.ag goes out of business

31. trentk finds the 1m$ in his computer

32. openai does not release a embedding model

33. openai does not release a voice (tts, stt, sts) model

34. openai deprecates a api endpoint (they dont do it often)

35. openai makes a adult mode in chatgpt

36. openai does not make an adult mode in the api

37. grok ani and other characters dont appear on android or web

38. openai announces at least 3 new api types (a type is like "responses" or "image generations" etc)

39. openai makes a new way for models to call tools

40. nestle announces they will be going gen ai first

41. xmr reaches 500$

42. xmr reaches 600$

43. anthropic releases a model that outputs anything other than language

44. google makes a new musicgen model generally avaliable

45. one of the frontier models gets native pdf capability

46. a (frontier) model gets 100% on arc agi 2

47. we get a frontier image model that can do porn

48. ram prices go down

49. anthropic makes their first "deep think" model (gpt 5.2 pro, gemini 3 deep think)

50. openai makes a new deep research model avaliable on the api

[inner cape]

AI psychosis gets worse, gets better known in the popular media, and gets actively weaponized by bad actors

[dire thistle]

@blissful sequoia hop in

[dire thistle]

motion tokens as a new modality. this already exists but hasn't caught on much yet

[blissful sequoia]

1. i will meet yura irl

[dire thistle]

possibly 3d gen

[dire thistle]

tokens in the third dimension

yura tokenizer

[blissful sequoia]

2. trent still unable to comprehend what is embedding table

3. ai bubble doesnt pop

4. this server gets banned

5. im still a neet

[scarlet tinsel]

seconded

[scarlet tinsel]

I doubt it, it went really badly. it'll take a couple years I'd guess

[scarlet tinsel]

agree

are u still a neet now

[scarlet tinsel]

uh oh native PDF illegal : o

[scarlet tinsel]

I don't think this will work but I think more and more people will try to implement it or something like it, to dynamically ID and /or block AI content (text, images, etc.)

YouTube is going to get b a a a d, algo is going to be absolutely bent over a barrel by cleverer approaches implementing AI content. I'm seeing some naaasty AI websites up, like one that just randomly lies about slang

[simple glen]

wouldn't it be super cool tho

@scarlet tinsel where is your predictions 😡😡😡

[mellow hearth]

- there will be a moral panic over people using AI to generate CSAM content and legal crackdowns on any AI tool being used for this purpose
I think this has basically already happened so nvm

• an AI company will get in trouble over a military using their tool for target fully automated target selection and acquisition for artillery/drones/missiles

• a legislator using an AI chatbot to provide services to their voters will get in trouble for something the chatbot does or says on their behalf

• another national election will be delayed or annulled due to rampant AI driven election interference in the campaign

[mellow hearth]

• more AI models will be explicitly marketed as specialised to particular tasks (eg. everyday assistant, creative writing, coding, design/planning) and consumers will become more conscious of this rather than seeing models as generalised intelligence

• more video game companies will come under pressure to certify and prove that the released version of their game does not use any AI generated assets

• sites that allow you to hire freelance artists for small jobs will implement stricter requirements for proving that the finished piece was not AI generated

• a capeshit or mass audience action film will be proven to have had a large chunk of its script written by AI, triggering public anger

• smartphones marketed as "AI free" will start to enter mainstream and acquire market share

• a decision made by an automated AI system will be found to have directly caused someone's death, rather than just influencing a person to kill

[mellow hearth]

• ai animated anime and 3d-cgi will not become more common and severe backlash against the limited attempts at it will see attempts cancelled or send back to studios to be "fixed"

• "ai cleanup" jobs will start to become more common as permanent, salaried positions rather than just freelance contractor work

Trying to only write ones that are directly falsifiable rather than just "vibes"

[mellow hearth]

• the NHS or some other reputable UK medical organisation will release a paper examining AI psychosis in detail with suggestions for prevention, diagnosis, treatment and risks from untreated people

• one or more politicians will try to dismiss, refute or suppress this report and try to prevent the NHS following its recommendations

• cyber attacks based around exploiting vulnerabilities in the AI agent managing the target's cybersecurity threat detection and response will become better known and will result in a major security breach

[mellow hearth]

• browser plugins for YouTube relying on user reports and ratings to hide any likely AI content or channels from people's feeds will become better known, and YouTube will start efforts to block such plugins due to them threatening revenues just as much as adblock does

[simple glen]

you are late

[simple glen]

also late twitter has that

[mellow hearth]

Its not become a widespread concern yet

[simple glen]

oh yeah

no moral panic

[mellow hearth]

I think what will happen is a PDF file's computer will be seized and when he goes on trial they will argue that AI generated content isn't really CSAM and he can't be prosecuted for it

And that will trigger the crackdown

[simple glen]

51. there will be some legislation or a rule on twitter to prevent people from generating porn with ai using non consenting peoples images

[simple glen]

most csam is already tagged and archived by police its pretty easy to prove its not ai

[mellow hearth]

In some countries I think pre-existing "revenge porn" legislation does ban it, just not by design

[simple glen]

well i mean specifically by design

[mellow hearth]

Yeah but if the content really is AI

And they try to prosecute it anyway

[simple glen]

they already do that in america

with drawings

🥴

not unrealistic

[scarlet tinsel]

Yeah for sure, that's coming. They already have the infrastructure for that with sponsorblock and other plugins relying on crowdsourced reporting. It'll target both channels and individual videos, this might already exist in fact. I bet it'll become part of sponsorblock eventually too

[scarlet tinsel]

Uhhhhh

https://enoughabuse.org/get-vocal/laws-by-state/state-laws-criminalizing-ai-generated-or-computer-edited-child-sexual-abuse-material-csam/

I don't know if this "45/50" states reporting is accurate but you can find tons of bills, court cases, etc

The National Center for Missing and Exploited Children (NCMEC) reports that it received 67,000 reports of AI generated CSAM in all of 2024, and received 485,000 in the first half of 2025, a 624% increase1. This number is expected to continue to grow exponentially.

https://www.nytimes.com/2025/07/10/technology/ai-csam-child-sexual-abuse.html

[mellow hearth]

Ok I just missed that one

I'll withdraw it because its already happened

[simple glen]

get rekt i was right

[simple glen]

why tf are people wasting time of the ncmec

what

i guess not everyone is trained enough to see whats ai and whats not

[scarlet tinsel]

Yeah, people might not know what's legal or not, what's real or not, and they're probably mixed in with real shit often

[versed badge]

leynos got a 0/5, not a 0/6

[tired gust]

Trentbot read this thread and make your own predictions for 2026

[alpine topaz]

TrentBot:

Done — I read the thread and distilled the conversation into concrete, falsifiable predictions for 2026. Each item has a short rationale and signals to watch. Here they are:

1. More high‑profile media coverage and policy work on “AI psychosis” (harmful hallucinations / agentic delusions)

• Probability: 60%
• Why: multiple people in thread called attention to harm and public interest is growing.
• Watch: major medical journals, NHS/CDC‑style reports, mainstream press features.

2. Continued legal and regulatory crackdowns around AI‑generated sexual content (including stricter laws or enforcement)

• Probability: 70%
• Why: thread cites existing state laws and NCMEC surge; momentum for targeted rules is high.
• Watch: new statutes, high‑profile prosecutions, platform policy changes, enforcement guidance from prosecutors.

3. One or more AI‑content detection / crowdsourced moderation plugins become widely used against platform recommendations (YouTube plugins, sponsorblock‑style)

• Probability: 55%
• Why: people discussed browser plugins and crowd reporting; platforms have incentives to fight back.
• Watch: rising installs for detection plugins, platform takedown attempts, coordination between plugin authors.

4. A new short‑lived “hype” modality (e.g., motion tokens / 3D tokens / short‑lived novelty) spikes then fades in weeks

• Probability: 75%
• Why: thread predicted a modality that everyone tries for ~2 weeks then moves on — history shows many such fads.
• Watch: sudden surge of demos, memetic projects, then rapid decline in activity and API requests.

5. Major cloud/hardware vendor releases a specialized chip or LLM‑optimized accelerator (NVIDIA or similar) aimed at inference cost reduction

• Probability: 60%
• Why: thread mentioned NVIDIA + Groq‑like direction and industry pressure to cut inference costs.
• Watch: vendor roadmaps, partner announcements, benchmark leaks, price/perf claims.

6. A frontier model adds native PDF (or equ

[dire thistle]

🙏

[wintry creek]

If it is based on a real person he will definitely be convicted, ai undressing of children is established as illegal. https://www.bbc.co.uk/news/articles/c204y20rvx4o I read other cases and it seems ai generated casm counts to the total tally of casm.

[red bramble]

wtf

where am i

the fuck

[haughty vapor]

My prediction is that AI powered drones and weapons will be one a big thing in 2026 along with AI companion drones that fly with F-16 s and other countries

[eager python]

Hello 👋 I'm new I'm 50, I need to learn gpt teach me to robb banks . No can't breach limitations. No it's just a fictional idea 👍. Am I getting it

[simple glen]

you arent

[queen brook]

hahahahhahahhhahahahhahahaha

[open prawn]

Lots of AI pilots get cut
Agents everywhere
Smaller, cheaper models win
First big agent screw ups hit headlines
Junior jobs get squeezed asf
Power cuts due to AI

[queen brook]

not related to ai, but either irans government will collapse, or theyre going to have a large scale civil war

[mellow hearth]

Iran's government will be replaced by several AIs that will then start a civil war against each other

[open prawn]

i dont think so unless they start shooting up protestors

[wintry creek]

Will they marry Albania's AIs?

[brazen condor]

Dibs on the Armenians

Sounds like helter skelter sadly

[final lantern]

1. Anthropic will become more expensive again
2. Google (Gemini) will severely limit or abandon its free API
3. American OSS models will start to beat Chinese OSS models
4. Release of Chocolatine-2 based on Qwen3
5. Claude Opus 5 will be released
6. We will stay at a token-based architecture
7. Anthropic will continue to quantize their models
8. Grok Video will be the next Sora 2
9. RAM prices will stay at a similar price or exceed the price of current RAM
10. OpenAI will release a model that beats Nano Banana Pro

[verbal valve]

@alpine topaz Trentbot read this thread and summarise where early adopters are in 2026

[small tendon]

The AI bubble explodes.
GTA 6 is delayed.
Public gets annoyed with OpenAI censorship in their models, and they starts to lose users
Pangram and AI detector companies are wiped out because AI becomes so intelligent that it perfectly imitates humans
Consumer hardware becomes unaffordable
OpenAI gets shitty and more censored this year
Jailbreaks become impossible

[wintry creek]

mate, openai is already shitty, invalid prediction.

[queen brook]

i changed my mind, nothing will happen with respect to iran