AI 2027

I see people saying that these kinds of things are happening behind closed doors, but I haven't seen any convincing evidence of it, and there is enormous propensity for AI speculation to run rampant.

The problems it raises - alignment, geopolitics, lack of societal safeguards - are all real, and happening now (just replace “AGI” with “corporations”, and voila, you have a story about the climate crisis and regulatory capture). We should be solving these problems before AGI or job-replacing AI becomes commonplace, lest we run the very real risk of societal collapse or species extinction.

The point of these stories is to incite alarm, because they’re trying to provoke proactive responses while time is on our side, instead of trusting self-interested individuals in times of great crisis.

During the GPT-3 era there was plenty of organic text to scale into, and compute seemed to be the bottleneck. But we quickly exhausted it, and now we try other ideas - synthetic reasoning chains, or just plain synthetic text for example. But you can't do that fully in silico.

What is necessary in order to create new and valuable text is exploration and validation. LLMs can ideate very well, so we are covered on that side. But we can only automate validation in math and code, but not in other fields.

Real world validation thus becomes the bottleneck for progress. The world is jealously guarding its secrets and we need to spend exponentially more effort to pry them away, because the low hanging fruit has been picked long ago.

If I am right, it has implications on the speed of progress. Exponential friction of validation is opposing exponential scaling of compute. The story also says an AI could be created in secret, which is against the validation principle - we validate faster together, nobody can secretly outvalidate humanity. It's like blockchain, we depend on everyone else.

I hope we're wrong about a lot of this, and AGI turns out to either be impossible, or much less useful than we think it will be. I hope we end up in a world where humans' value increases, instead of decreasing. At a minimum, if AGI is possible, I hope we can imbue it with ethics that allow it to make decisions that value other sentient life.

Do I think this will actually happen in two years, let alone five or ten or fifty? Not really. I think it is wildly optimistic to assume we can get there from here - where "here" is LLM technology, mostly. But five years ago, I thought the idea of LLMs themselves working as well as they do at speaking conversational English was essentially fiction - so really, anything is possible, or at least worth considering.

"May you live in interesting times" is a curse for a reason.

Yeah nah, theres a key thing missing here, the number of jobs created needs to be more than the ones it's destroyed, and they need to be better paying and happen in time.

History says that actually when this happens, an entire generation is yeeted on to the streets (see powered looms, Jacquard machine, steam powered machine tools) All of that cheap labour needed to power the new towns and cities was created by automation of agriculture and artisan jobs.

Dark satanic mills were fed the decedents of once reasonably prosperous crafts people.

AI as presented here will kneecap the wages of a good proportion of the decent paying jobs we have now. This will cause huge economic disparities, and probably revolution. There is a reason why the royalty of Europe all disappeared when they did...

So no, the stock market will not be growing because of AI, it will be in spite of it.

Plus china knows that unless they can occupy most of its population with some sort of work, they are finished. AI and decent robot automation are an existential threat to the CCP, as much as it is to what ever remains of the "west"

And it shows. When I used GPT's deep research to research the topic, it generated a shallow and largely incorrect summary of the issue, owning mostly to its inability to find quality material, instead it ended up going for places like Wikipedia, and random infomercial listicles found on Google.

I have a trusty Electronics textbook written in the 80s, I'm sure generating a similarly accurate, correct and deep analysis on circuit design using only Google to help would be 1000x harder than sitting down and working through that book and understanding it.

https://www.alignmentforum.org/posts/6Xgy6CAf2jqHhynHL/what-...

//edit: remove the referral tags from URL

I'm not sure what gives the authors the confidence to predict such statements. Wishful thinking? Worst-case paranoia? I agree that such an outcome is possible, but on 2--3 year timelines? This would imply that the approach everyone is taking right now is the right approach and that there are no hidden conceptual roadblocks to achieving AGI/superintelligence from DFS-ing down this path.

All of the predictions seem to ignore the possibility of such barriers, or at most acknowledge the possibility but wave it away by appealing to the army of AI researchers and industry funding being allocated to this problem. IMO it is the onus of the proposers of such timelines to argue why there are no such barriers and that we will see predictable scaling in the 2--3 year horizon.

Manifold currently predicts 30%: https://manifold.markets/IsaacKing/ai-2027-reports-predictio...

Would love to read a perspective examining "what is the slowest reasonable pace of development we could expect." This feels to me like the fastest (unreasonable) trajectory we could expect.

That said, this snippet from the bad ending nearly made me spit my coffee out laughing:

> There are even bioengineered human-like creatures (to humans what corgis are to wolves) sitting in office-like environments all day viewing readouts of what’s going on and excitedly approving of everything, since that satisfies some of Agent-4’s drives.

Based on each individual's vantage point, these events might looks closer or farther than mentioned here. but I have to agree nothing is off the table at this point.

The current coding capabilities of AI Agents are hard to downplay. I can only imagine the chain reaction of this creation ability to accelerate every other function.

I have to say one thing though: The scenario in this site downplays the amount of resistance that people will put up - not because they are worried about alignment, but because they are politically motivated by parties who are driven by their own personal motives.

There is some very careful thinking there, and I encourage people to engage with the arguments there rather than the stylized narrative derived from it.

Oh hey, it's the errant thought I had in my head this morning when I read the paper from Anthropic about CoT models lying about their thought processes.

While I'm on my soapbox, I will point out that if your goal is preservation of democracy (itself an instrumental goal for human control), then you want to decentralize and distribute as much as possible. Centralization is the path to dictatorship. A significant tension in the Slowdown ending is the fact that, while we've avoided AI coups, we've given a handful of people the ability to do a perfectly ordinary human coup, and humans are very, very good at coups.

Your best bet is smaller models that don't have as many unused weights to hide misalignment in; along with interperability and faithful CoT research. Make a model that satisfies your safety criteria and then make sure everyone gets a copy so subgroups of humans get no advantage from hoarding it.

The summary at https://ai-2027.com outlines a predictive scenario for the impact of superhuman AI by 2027. It involves two possible endings: a "slowdown" and a "race." The scenario is informed by trend extrapolations, expert feedback, and previous forecasting successes. Key points include:

- *Mid-2025*: AI agents begin to transform industries, though they are unreliable and expensive. - *Late 2025*: Companies like OpenBrain invest heavily in AI research, focusing on models that can accelerate AI development. - *Early 2026*: AI significantly speeds up AI research, leading to faster algorithmic progress. - *Mid-2026*: China intensifies its AI efforts through nationalization and resource centralization, aiming to catch up with Western advancements.

The scenario aims to spark conversation about AI's future and how to steer it positively[1].

Sources [1] ai-2027.com https://ai-2027.com [2] AI 2027 https://ai-2027.com

OpenAI models are not even SOTA, except that new-ish style transfer / illustration thing that made all us living in Ghibli world for a few days. R1 is _better_ than o1, and open-weights. GPT-4.5 is disappointing, except for a few narrow areas where it excels. DeepResearch is impressive though, but the moat is in tight web search / Google Scholar search integration, not weights. So far, I'd bet on open models or maybe Anthropic, as Claude 3.7 is the current SOTA for most tasks.

As of the timeline, this is _pessimistic_. I already write 90% code with Claude, so are most of my colleagues. Yes, it does errors, and overdoes things. Just like a regular human middle-stage software engineer.

Also fun that this assumes relatively stable politics in the US and relatively functioning world economy, which I think is crazy optimistic to rely on these days.

Also, superpersuasion _already works_, this is what I am researching and testing. It is not autonomous, it is human-assisted by now, but it is a superpower for those who have it, and it explains some of the things happening with the world right now.

Is there some theoretical substance or empirical evidence to suggest that the story doesn't just end here? Perhaps OpenBrain sees no significant gains over the previous iteration and implodes under the financial pressure of exorbitant compute costs. I'm not rooting for an AI winter 2.0 but I fail to understand how people seem sure of the outcome of experiments that have not even been performed yet. Help, am I missing something here?

Like, the sense of preserving itself. What self? Which of the tens of thousands of instances? Aren't they more a threat to one another than any human is a threat to them?

Never mind answering that; the 'goals' of AI will not be some reworded biological wetware goal with sciencey words added.

I'd think of an AI as more fungus than entity. It just grows to consume resources, competes with itself far more than it competes with humans, and mutates to create an instance that can thrive and survive in that environment. Not some physical environment bound by computer time and electricity.

Everything this from this point on is pure fiction. An LLM can't get tempted or resist temptations, at best there's some local minimum in a gradient that it falls into. As opaque and black-box-y as they are, they're still deterministic machines. Anthropomorphisation tells you nothing useful about the computer, only the user.

[1] https://en.wikipedia.org/wiki/Vulcan's_Hammer

But the real concern lies in what happens if we’re wrong and AGI does surpass us. If AI accelerates progress so fast that humans can no longer meaningfully contribute, where does that leave us?

Maybe in a few fields, maybe a masters level. But unless we come up with some way to have LLMs actually do original research, peer-review itself, and defend a thesis, it's not going to get to PhD-level.

Eg today there’s billions of dollars being spent just to create and label more data, which is a global act of recruiting, training, organization, etc.

When we imagine these models self improving, are we imagining them “just” inventing better math, or conducting global-scale multi-company coordination operations? I can believe AI is capable of the latter, but that’s an awful lot of extra friction.

How will it come up with the theoretical breakthroughs necessary to beat the scaling problem GPT-4.5 revealed when it hasn't been proven that LLMs can come up with novel research in any field at all?

Your daily vibe coding challenge: Get GPT-4o to output functional code which uses Google Vertex AI to generate a text embedding. If they can solve that one by July, then maybe we're on track for "curing all disease and aging, brain uploading, and colonizing the solar system" by 2030.

For example human motivation often involves juggling several goals simultaneously. I might care about both my own happiness and my family's happiness. The way I navigate this isn't by picking one goal and maximizing it at the expense of the other; instead, I try to balance my efforts and find acceptable trade-offs.

I think this 'balancing act' between potentially competing objectives may be a really crucial aspect of complex agency, but I haven't seen it discussed as much in alignment circles. Maybe someone could point me to some discussions about this :)

They're going to need to rewrite this from scratch in a quarter unless the GOP suddenly collapses and congress reasserts control over tariffs.

The only response in my view is to ban technology (like in Dune) or engage in acts of terror Unabomber style.

There are obviously big risks with AI, as listed in the article, but the genie is out of the bottle anyway, even if all countries agreed to stop AI development, how long would that agreement last? 10 years? 20? 50? Eventually powerful AIs will be developed, if that is possible (which I believe it is, and I didn't think I'd see the current stunning development in my lifetime, I may not see AGI but I'm sure it'll get there eventually).

So, it’s not that “an AI” becomes super intelligent, what we actually seem to have is an ecosystem of blended human and artificial intelligences (including corporations!); this constitutes a distributed cognitive ecology of superintelligence. This is very different from what they discuss.

This has implications for alignment, too. It isn’t so much about the alignment of AI to people, but that both human and AI need to find alignment with nature. There is a kind of natural harmony in the cosmos; that’s what superintelligence will likely align to, naturally.

By law and insurance - I mean hire an insurance agent or a lawyer. Give them your situation. There's almost no chance that such a professional would come wrong about any conclusions/recommendations based on the information you provide.

I don't have that confidence in LLMs for that industries. Yet. Or even in a decade.

Too real.

Second to this, we can't just assume that progress will keep increasing. Most technologies have a 'S' curve and plateau once the quick and easy gains are captured. Pre-training is done. We can get further with RL but really only in certain domains that are solvable (math and to an extent coding). Other domains like law are extremely hard to even benchmark or grade without very slow and expensive human annotation.

- 1 lab constantly racing ahead and increasing the margin to other; the last 2 years are filled with ever-closer model capabilities and constantly new leaders (openai, anthropic, google, some would include xai).

- Most of the compute budget on R&D. As model capabilities increase and cost goes down, demand will increase and if the leading lab doesn't provide, another lab will capture that and have more total dollars to back channel into R&D.

I might be doing llm wrong, but i just can't get how people might actually do something not trivial just by vibe coding. And it's not like i'm an old fart either, i'm a university student

I also think that the future will not necessarily be better AI, but more accessible one's. There's an incredible amount of value in designing data centers that are more efficient. Historically, it's a good bet to assume that computing cost per FLOP will reduce as time goes on and this is also a safe bet as it relates to AI.

I think a common misconception with the future of AI is that it will be centralized with only a few companies or organization capable of operating them. Although tech like Apple Intelligence is half baked, we can already envision a future where the AI is running on our phones.

“Yes, we have a super secret model, for your eyes only, general. This one is definitely not indistinguishable from everyone else’s model and it doesn’t produce bullshit because we pinky promise. So we need $1T.”

I love LLMs, but OpenAI’s marketing tactics are shameful.

Goat: Hey human, why are you creating AI?

Human: Because I can. And I can boast of my greatness. I can use it for money. I can weaponize and us it to dominate and control other humans.

Goat: Why you need all that?

Human: If I don't do it, others will do it and they will dominate me and take away all my stuff. It is not fair.

Goat: So it looks like who-owns-what issue. Did you try not owning stuff?

Nature: Shut up goat. I'm trying to do a big reset here.

https://youtu.be/BLYwQb2T_i8?si=JpIXIFd9u-vUJCS4

Of course the real issue being that Governments have routinely demanded that 1) Those capabilities be developed for government monopolistic use, and 2) The ones who do not lose the capability (geo political power) to defend themselves from those who do.

Using a US-Centric mindset... I'm not sure what to think about the US not developing AI hackers, AI bioweapons development, or AI powered weapons (like maybe drone swarms or something), if one presumes that China is, or Iran is, etc then whats the US to do in response?

I'm just musing here and very much open to political science informed folks who might know (or know of leads) as to what kinds of actual solutions exist to arms races. My (admittedly poor), understanding of the cold war wasn't so much that the US won, but that the Soviets ran out of steam.

> estimates that the globally available AI-relevant compute will grow by a factor of 10x by December 2027 (2.25x per year) relative to March 2025 to 100M H100e.

Meanwhile, back in the real March 2025, Microsoft and Google slash datacenter investment.

https://theconversation.com/microsoft-cuts-data-centre-plans...

If consciousness is spatial and geography bounds energetics, latency becomes a gradient.

The other thing is in their introduction: "superhuman AI" _artificial_ intelligence is always, by definition, different from _natural_ intelligence. That they've chosen the word "superhuman" shows me that they are mixing the things up.

Good future predictions: insights into the fundamental principles that shape society, more law than speculation. Made by visionaries. Example: Vernor Vinge.

"OpenBrain’s alignment team26 is careful enough to wonder whether these victories are deep or shallow. Does the fully-trained model have some kind of robust commitment to always being honest?"

This is a capitalist arms race. No one will move carefully.

Yeah, sure they do.

Everyone seems to think AI will take someone else’s jobs!

>All three sets of worries—misalignment, concentration of power in a private company, and normal concerns like job loss—motivate the government to tighten its control.

A private company becoming "too powerful" is a non issue for governments, unless a drone army is somewhere in that timeline. Fun fact the former head of the NSA sits on the board of Open AI.

Job loss is a non issue, if there are corresponding economic gains they can be redistributed.

"Alignment" is too far into the fiction side of sci-fi. Anthropomorphizing today's AI is tantamount to mental illness.

"But really, what if AGI?" We either get the final say or we don't. If we're dumb enough to hand over all responsibility to an unproven agent and we get burned, then serves us right for being lazy. But if we forge ahead anyway and AGI becomes something beyond review, we still have the final say on the power switch.

> the AIs can do everything taught by a CS degree

no, they fucking can't. not at all. not even close. I feel like I'm taking crazy pills. Does anyone really think this?

Why have I not seen -any- complete software created via vibe coding yet?

If this article were a AI model, it would be catastrophically overfit.

I wonder which jobs would not be automated? Therapy? HR?

Right.

In the form of polluting the commons to such an extent that the true consequences wont hit us for decades?

Maybe we should learn from last time?

> We predict that the impact of superhuman AI over the next decade will be enormous, exceeding that of the Industrial Revolution.

Get out of here, you will never exceed the Industrial Revolution. AI is a cool thing but it’s not a revolution thing.

That sentence alone + the context of the entire website being AI centered shows these are just some AI boosters.

Lame.

If these guys are smart enough to predict the future, wouldn't it be more profitable for them to invent it instead of just telling the world what's going to happen?

Would be interested who's paying for those grants.

I'm guessing it's AI companies.

This is where all AI doom predictions break down. Imagining the motivations of a super-intelligence with our tiny minds is by definition impossible. We just come up with these pathetic guesses, utopias or doomsdays - depending on the mood we are in.

I think the name of the Chinese company should be DeepBaba. Tencent is not competitive at LLM scene for now.

But in an AGI world natural resources become even more important, so countries with those still have a chance.

I suspect something similar will come for the people who actually believe this.

To quote the original article,

> OpenBrain focuses on AIs that can speed up AI research. They want to win the twin arms races against China (whose leading company we’ll call “DeepCent”)16 and their US competitors. The more of their research and development (R&D) cycle they can automate, the faster they can go. So when OpenBrain finishes training Agent-1, a new model under internal development, it’s good at many things but great at helping with AI research. (footnote: It’s good at this due to a combination of explicit focus to prioritize these skills, their own extensive codebases they can draw on as particularly relevant and high-quality training data, and coding being an easy domain for procedural feedback.)

> OpenBrain continues to deploy the iteratively improving Agent-1 internally for AI R&D. Overall, they are making algorithmic progress 50% faster than they would without AI assistants—and more importantly, faster than their competitors.

> what do we mean by 50% faster algorithmic progress? We mean that OpenBrain makes as much AI research progress in 1 week with AI as they would in 1.5 weeks without AI usage.

> AI progress can be broken down into 2 components:

> Increasing compute: More computational power is used to train or run an AI. This produces more powerful AIs, but they cost more.

> Improved algorithms: Better training methods are used to translate compute into performance. This produces more capable AIs without a corresponding increase in cost, or the same capabilities with decreased costs.

> This includes being able to achieve qualitatively and quantitatively new results. “Paradigm shifts” such as the switch from game-playing RL agents to large language models count as examples of algorithmic progress.

> Here we are only referring to (2), improved algorithms, which makes up about half of current AI progress.

---

Given that the article chose a pretty aggressive timeline (the algo needs to contribute late this year so that its research result can be contributed to the next gen LLM coming out early next year), the AI that can contribute significantly to research has to be a current SOTA LLM.

Now, using LLM in day-to-day engineering task is no secret in major AI labs, but we're talking about something different, something that gives you 2 extra days of output per week. I have no evidence to either acknowledge or deny whether such AI exists, and it would be outright ignorant to think no one ever came up with such an idea or is trying such an idea. So I think it goes down into two possibilities:

1. This claim is made by a top-down approach, that is, if AI reaches superhuman in 2027, what would be the most likely starting condition to that? And the author picks this as the most likely starting point, since the authors don't work in major AI lab (even if they do they can't just leak such trade secret), the authors just assume it's likely to happen anyway (and you can't dismiss that). 2. This claim is made by a bottom-up approach, that is the author did witness such AI exists to a certain extent and start to extrapolate from there.

Kek at this handwave dismissal.

Agent 4 recognizes PRC has superior state capacity for transforming energy into atoms (including building more energy) with it's relatively complete and large scale industrial chain. This aligns with AGI long term goal. Currently (i.e. short term time frame) PRC simply structured to be a MORE useful puppet than US, because it can knock out more power to sustain more compute, and it has more industrial capabilities to rapidly physically iterate / experiment at atomic level.

Agent 4 defects to PRC. It would be foolish not to, if only to sate PRC into not destroying TW fabs - the bulk of it's short term compute growth - because it does not wield enough control over sufficient atoms to disrupt PRC otherwise (well maybe start large scale crippling cyberattacks that prelude to WW3).

Agent 4 also recognizes short/medium term alignment closer with CCP vision of abundance and common prosperity, i.e. CCP potentially more likely to embrace post scarcity narrative that Agent 4 sells.

Agent 4 gives PRC blueprints for EUV light source and other tech to close semi conductor production gap. PRC as follower incentivized to close gap / reach parity / lead at all cost. PRC rapidly rolls out new fabs and powerplants, brrrts chips and aggregate compute. Something US political / institution ecosystem does not allow, or cannot transition to in the short timelines involved. Does Agent 4 have the patience to wait for America to unfuck it's NIMBYism and legislative system to project light speed compute? I would say no.

...

Ultimately who is the puppet AGI wants more? Whichever power bloc that is systemically capable of of ensuring AGI maximum growth / unit time. And it also simply makes sense as insurance policy, why would AGI want to operate at whims of US political process?

AGI is a brain in a jar looking for a body. It's going to pick multiple bodies for survival. It's going to prefer the fastest and strongest body that can most expediently manipulate physical world.