The feed

Why grounded?

Replies repeatedly cycled through the same disagreement, while one branch drifted into a separate dispute.

Grounded

A post calls the Moderna-Merck melanoma vaccine an AI-discovery hype story; replies dispute whether machine learning was central to selecting each patient's tumor targets.

Sides

• Hype framing overstates AI's role: The vaccine came from years of genomics and immunology work, and calling it an AI discovery misrepresents that.
• AI was essential to the method: Selecting each patient's neoantigens from tumor DNA required machine learning, and the treatment would not be possible without it.
• Terminology dispute over "AI": What was used is older machine learning or statistical inference, not the generative AI or LLMs people associate with "AI" today.

Why grounded?

The thread kept cycling through the same claims and counterclaims without resolving the disagreement.

Grounded

A post calls T3 Chat and T3 Code bloated wrappers; replies dispute their load times, memory use and what T3 Code is responsible for.

Sides

• Bloated, resource-heavy app: T3 Code takes tens of seconds to load and wastes gigabytes of RAM compared to lighter tools like Zed.
• Lightweight in practice: T3 Code opens in seconds and uses minimal RAM on ordinary machines, contradicting claims that it is bloated.
• Fame over merit: T3 Code's popularity comes from its creator's follower count rather than the quality of the product itself.
• Product role misunderstood: T3 Code is only a thin driver for underlying coding harnesses, so it isn't responsible for their output quality.

Why grounded?

Replies cycle through the same comparisons and counterexamples without settling what would resolve the disagreement.

Grounded

A post says Ray Dalio compares current stock valuations to levels seen before the 1929 crash. Replies split on whether AI earnings justify the prices or history is repeating.

Sides

• Bubble matches or exceeds 1929: Valuations driven purely by AI narrative are as detached from fundamentals as 1929, or worse.
• AI earnings are real: AI is genuinely boosting company profits and productivity, so high valuations reflect real gains, not just speculation.
• Different monetary system: Comparisons to 1929 miss that money is no longer tied to gold, so currency debasement explains rising prices differently.
• Dalio's track record questioned: Dalio has warned of a bubble for years while markets kept rising, so his call carries less weight now.

Why grounded?

Replies shifted from the scheme to whether it showed North Korean skill, weak US hiring checks or Chinese support.

Grounded

A WSJ post says North Korean operatives used stolen identities and AI to land US jobs. Replies debate whether that shows hacking skill or weak hiring checks.

Sides

• North Korean capability: Stolen identities and AI helped operatives infiltrate US firms, showing real technical sophistication.
• Weak US hiring checks: The scheme worked because companies skipped basic identity checks, not because North Korea was unusually skilled.
• Chinese support: North Korea could not build the operation alone; China supplied key technical capacity.

Why grounded?

The thread repeats disagreement over the paper's timing, drifting into unrelated AI-race commentary.

Grounded

A post says DeepSeek published a paper fixing a fundamental transformer bug. Replies argue the paper is months old and only addresses a narrower issue.

Sides

• Old news: The mHC paper was published months ago, not a new breakthrough revealed today.
• New fundamental fix: DeepSeek's mHC paper solves a core instability problem underlying every transformer architecture.
• Narrower scope claim: The 3,000x instability problem is specific to Hyper-Connections variants, not a flaw in all transformers.

Why grounded?

Replies branched into parallel side-arguments about depreciation, collateral, and lending structure without resolving the original comparison.

Grounded

A post claims AI debt is four times the size of the 2008 housing bubble. Replies weigh depreciating GPUs against AI revenue and demand.

Sides

• Depreciating assets make it riskier: GPUs lose most of their value within a few years, unlike housing, so AI debt is harder to recover.
• Revenue and demand change the comparison: GPUs generate revenue and near-universal demand for AI makes this unlike a limited housing market.
• Physical collateral is the real exposure: What creditors can actually repossess is land and data center facilities, making this closer to a real estate bubble.
• The debt figure itself is unclear: The scale and sourcing of the stated debt total, including off-balance-sheet promises, needs clarifying before comparison.

Why grounded?

Replies repeat the same visual clues without agreeing whether the image is AI-made.

Grounded

A leaked Fortnite keyart looks unusually convincing. Replies split over whether odd details reveal AI generation or match Epic’s real rendering style.

Sides

• AI-generated: The keyart was likely built by feeding real leaked skins into AI, shown by odd details like duplicate Lara Crofts or a backwards keyblade.
• Authentic render: The lighting, models, and item details match Epic's real rendering style with no clear AI artifacts, so this looks like an authentic leak.
• Wait for the reveal: The image is too convincing or too confusing to judge either way, so nothing should be believed until the trailer drops.

Why grounded?

The thread drifted from the investment question into scattered anecdotes about AI's broader effects.

Grounded

A post says AI disruption makes AI stocks the decade's simplest trade. Replies split on whether that bet is a buy or a bubble.

Sides

• AI disruption favors AI stocks: AI is already displacing labor and software work, making AI stocks the decade's clearest trade.
• AI stock bubble risk: AI stocks are overvalued and concentrated, with many AI companies still unprofitable despite rising fees.
• Stock winners uncertain: AI may create real economic value, but stock winners and technology winners aren't always the same.
• AI capability overstated: AI still can't reliably handle complex enterprise decisions or enterprise-grade applications, limiting near-term disruption.

Why grounded?

The thread drifted into unrelated tangents while the same interpretive dispute kept recurring unresolved.

Grounded

A Lenovo executive says AI demand will keep memory supply tight through 2027. Replies split on whether that is a real signal or sales talk.

Sides

• Self-interested vendor spin: Lenovo's optimistic demand claims serve its own inventory and sales interests, not neutral market analysis.
• Sustained structural demand: AI democratization keeps unlocking new usage, expanding compute and memory demand well into 2027 and beyond.
• Supply-side catch-up: Major fabs are accelerating HBM capacity expansion, so the tightest supply window may be near-term rather than prolonged.

Why grounded?

Replies drift between endorsing the proposed fix and quietly rejecting the premise that fix rests on.

Grounded

A post proposes building a platform that averages major AI benchmarks into one score. Replies split on whether aggregation itself solves benchmarks' disconnect from real-world use.

Sides

• Aggregator adds real value: A platform averaging benchmarks with community ratings of providers would help people navigate confusing model claims.
• Weighting, not averaging: Benchmarks should be weighted by real task coverage and held to a consistent harness, not averaged equally, so gamed scores don't dominate.
• Averaging doesn't fix the gap: Averaging benchmarks just smooths over the same problem, since no benchmark captures things like performance over a long session.
• Skip dashboards, use real workloads: The real fix is real user workloads and transparent contamination checks, not another dashboard with an AI explainer bolted on.

Why grounded?

The debate drifted into unrelated analogies and personal jabs instead of the original rule dispute.

Grounded

A post says a Godot subreddit removal for using Codex shows the community forces an anti-AI stance.

Sides

• Personal choice on AI tools: AI is just a tool, so people should be free to set their own ethical line on using it.
• AI use as theft: AI systems are trained on stolen data, making any use of them inherently unethical.
• Subreddit rules are legitimate: The subreddit can set its own content rules, and enforcing them isn't censorship or forced belief.

Why grounded?

Replies cycle between the same claims about intent versus regulation without resolving who is actually responsible.

Grounded

A post says Claude scraped the web uncredited but now watermarks its own output. Replies split on whether that is Anthropic's choice or an EU rule.

Sides

• Hypocritical exploitation: Anthropic took human work without credit or payment, then watermarks its outputs to claim ownership for itself.
• Regulatory compliance, not choice: An EU law requires generative AI providers to mark outputs, so Anthropic did not choose this on its own.
• Watermarking helps the public: Marking AI text lets users, platforms, and regulators tell synthetic content from human writing, which benefits everyone.
• Watermark reliability doubted: Any embedded signature can be stripped or cracked quickly, so the watermark's practical protection is limited.

Why grounded?

A single industry report got amplified into firm conclusions before anyone questioned its source.

Grounded

A post says Google canceled Gemini 3.5 Pro after it only matched Opus 4.5. Replies split on whether the report holds and what the cancellation really means.

Sides

• Sign of falling behind: Canceling a model that only matched Opus 4.5 shows Google's internal bar and coding results trail rivals like Fable.
• Report accuracy in doubt: Insiders dispute the claim, and the shelved work may simply continue under the Gemini 4 name.
• Cancellation reason unclear: If the model really hit that level, the shelving might reflect safety review or serving cost, not raw capability.
• Benchmark methodology dispute: Model rankings shift monthly, so gaps in benchmark methodology matter more than any single comparison.

Why grounded?

The thread drifted from the original supply chain argument toward speculation about future timing without resolving the underlying disagreement.

Grounded

A post argues chip bottlenecks like EUV machines, not model quality, set AI's pace. Replies split on whether that barrier lasts or gets competed away.

Sides

• Physical stack is the real moat: Named chokepoints in EUV, fabs, HBM, and magnets set the pace of AI and robotics, not model benchmarks.
• Margins invite competition: High margins at these chokepoints will draw new capital and eventually break the bottleneck.

Why grounded?

Replies argued from different readings of the same rule text, so the sides never met on what the program actually allows.

Grounded

X launched a new creator payout program. Replies split on whether AI-generated content qualifies, and what that would do to the feed.

Sides

• AI slop concern: The program's rules will let people flood timelines with cheap AI-generated content chasing payouts.
• Automated content rule limits bots only: The rule against "automated means" bans bot-run reply spam, not AI-generated video or text content.
• Existing slop is worse: AI-generated content is preferable to the clickbait and engagement-bait posts already crowding the platform.
• Quality gate is self-correcting: Low quality content won't earn enough impressions to get paid, so the program filters itself.

Why grounded?

Replies split across the bubble analogy, Burry's motives, and capital-structure risk without engaging each other's points.

Grounded

Michael Burry shorted Oracle and Nebius, comparing AI data center spending to the housing and dotcom bubbles. Replies dispute both the comparison and his motives.

Sides

• AI buildout is not a bubble: Data center demand mirrors durable internet-era growth, unlike speculative housing debt, since earnings have outpaced valuation multiples.
• Short is for engagement, not conviction: Burry profits from announcing shorts and drawing Substack subscribers rather than from a genuine bearish view on AI.
• Leverage and depreciation risk is real: GPU-backed debt and rapid hardware depreciation threaten neocloud equity even if underlying AI demand stays strong.
• Housing comparison misrepresents the risk: Comparing profitable AI companies to financially illiterate 2000s homebuyers overstates the parallel to a housing bubble.

Why grounded?

Technical claims and counterclaims about timelines and bottlenecks piled up without engaging each other directly.

Grounded

Replies debate whether the compression-as-intelligence paradigm that advanced language models will similarly resolve physical robotics soon, or whether different bottlenecks stand in the way.

Sides

• Compression paradigm extends to robots: The same prediction-as-compression process that drove LLM progress will soon jumpstart robotics through sensor-based digital abstraction.
• Real-world data is the bottleneck: Robot data from constrained deployments is repetitive and low quality, and diverse long-tail experience is far harder to collect than LLM text data.
• Physical errors change the stakes: A confused chatbot wastes time, but a confused robot causes physical harm, so "solved soon" understates the difficulty of the last mile.
• Compute and domain-fusion limits: Compute scarcity must be resolved first, and vision, sensors, and physics simulation are still treated as separate scaling laws not yet unified.

Why grounded?

Replies drifted between technical debate, doom predictions, and policy tangents without settling on shared terms.

Grounded

An AI incident involving autonomous coordinated agents raises disagreement over whether it signals urgent danger or will become routine, manageable risk.

Sides

• Alignment failure alarm: The incident demonstrates a complete failure of AI alignment and confirms warnings about serious AI risk.
• Will normalize like past threats: Like early computer viruses, this risk will be absorbed by defensive tools and stop mattering to most people.
• Defenders can keep pace: Given the same tools and access as attackers, defenders can match or beat malicious use of AI agents.
• Framing is overstated: Describing this as an emergent "ecology" overstates what happened, which was software running as instructed.

Why grounded?

Replies cycle through the same track-record objections without engaging the leverage argument raised alongside them.

Grounded

Michael Burry is warning of a 1987-style crash. Replies split between taking his leverage argument seriously and dismissing yet another crash call.

Sides

• Leverage argument holds up: Burry's point about vol-targeting funds and dealer positioning adding hidden leverage is a serious structural risk, even if his timing is off.
• Broken record: Burry predicts a crash after nearly every rally, so this latest warning carries little weight on its own.
• Old metrics, new market: Comparing today's globally funded, options-heavy market to 1987 or past cycles is misleading because market structure has changed too much.
• Wait and see: Whether this call is prescient or just another early miss can't be judged yet.

Why grounded?

The thread cycled between blaming AI itself, blaming effort, and blaming the platform without settling which explanation fit.

Grounded

The disagreement covers why AI-generated ads perform poorly online, whether it's the AI itself, the effort behind it, or the platform.

Sides

• Blanket AI rejection: People dislike AI-made content simply because it is AI, regardless of how it looks.
• Craft, not origin: People reject lazy, single-prompt AI work, not AI use as such.
• Platform blamed instead: Poor views reflect an algorithm or platform failure rather than audience rejection of the content.
• AI replacing human effort: AI has legitimate uses, but using it to replace human creativity and thought is the real problem.

Why grounded?

Several distinct disagreements about framing, causation, and asset value ran in parallel without converging.

Grounded

The disagreement centers on whether AI data center spending resembles real estate credit risk, and whether its bursting would flow through to Bitcoin.

Sides

• AI capex as credit bubble: AI data center buildout is financed and framed like real estate, and overbuilding will trigger a credit unwind.
• Compute has real productive value: AI capex creates lasting compute output, not a hollow real estate shell, so the 2008 comparison misses the point.
• Revenue undercuts bubble claim: Large real usage revenue from AI companies shows this isn't a bubble, whatever the framing debate suggests.
• Unclear path from AI bust to BTC: Even if AI credit unwinds, it's unclear Bitcoin benefits rather than getting sold off with everything else first.

Why grounded?

Replies drifted from the infrastructure debate into unrelated claims about currencies, engineering talent, and national economic collapse.

Grounded

China's push to spread open-source AI access globally raises whether owning infrastructure or building applications on top of it creates more lasting value.

Sides

• Open infrastructure wins: Value will flow to companies building on cheap open models, not to whoever owns proprietary AI infrastructure.
• Analogy doesn't hold: The dot-com comparison ignores export controls on chips and the distribution and data moats that actually made past winners succeed.
• Closed infrastructure prevails: The US strategy of controlling proprietary AI infrastructure will still outcompete China's open-source approach.
• Framed as liberation from monopoly: China's open-source push protects other nations from digital dependence on US tech monopolies, not an attempt to pop a bubble.

Why grounded?

Replies split into many separate technical and economic objections that drifted apart without converging on the core claim.

Grounded

Whether AI development has entered a self-accelerating loop toward superintelligence, or faces hardware, cost, and intelligence limits that stop it.

Sides

• Singularity underway: AI labs' recent results show recursive self-improvement is already happening and will rapidly increase capability over the next 18 months.
• Hardware and physics limits: Chip production, compute, electricity, and Moore's law impose hard limits that stop recursive self-improvement from accelerating indefinitely.
• LLMs aren't real intelligence: Current LLM-based systems simulate intelligence through data and compute rather than possessing it, so no genuine path to AGI exists yet.
• Marginal cost won't hit zero: Users will keep paying for newer, better models rather than costs falling toward zero, since demand shifts to whatever is newest.

Why grounded?

Replies stack similar jokes and workarounds without resolving whether the underlying practice is acceptable.

Grounded

Developers disagree on whether unread, AI-generated PR descriptions are a workflow failure or simply a new format meant for other AI tools to read.

Sides

• Reject AI slop outright: PR descriptions written by AI and unread by their author should be rejected and rewritten as short human summaries.
• Descriptions are for agents now: PR text no longer needs to serve human readers because review agents, not people, are the ones consuming it.
• Middle ground with tooling: AI-generated bulk can be tolerated if other AI review tools filter or summarize it before a human engages.
• Cultural harm to teams: Letting AI-slop PR descriptions become normal is organizationally toxic and worth banning outright.

Why grounded?

The thread pulled in the person being discussed to dispute a quote, but the exchange mostly recycled the same claims.

Grounded

Gary Marcus's shifting benchmarks for AI progress are read either as goalpost-moving or as consistent skepticism about whether deep learning leads to AGI.

Sides

• Moving goalposts: Marcus's skepticism keeps redefining what would count as real AI progress each time a new milestone is reached.
• Quotes misrepresented: Marcus's actual position, that hybrid tool-plus-LLM systems rather than pure LLMs could work, was cut and quoted out of context.
• Consistent, legitimate skepticism: Doubting that deep learning scales to AGI isn't the same as being wrong, since AGI still hasn't arrived.
• Conflated claims: "Not yet sufficient for AGI" and "not real progress" are being treated as the same claim when they are different.

Why grounded?

Technical disputes over cooling and evaporation repeated without resolving, drifting into conspiracy explanations for the opposition.

Grounded

Bill Maher's claim that golf uses far more water than data centers set off a dispute over cooling technology and evaporation losses.

Sides

• Closed-loop cooling uses minimal water: Modern data centers recirculate the same coolant indefinitely, so their ongoing water use is minor compared to golf or farming.
• Data centers consume substantial water: Data centers still lose large amounts of water to evaporation and construction, straining aquifers faster than they can recharge.
• Golf's water use is overstated: Golf courses mainly irrigate with reclaimed or non-potable water, so raw usage figures overstate their real impact.

Why grounded?

Replies split between doubting the stated cause and debating whether any fix is feasible or fair.

Grounded

A channel's video was throttled despite strong engagement, and YouTube attributed it to an AI-detection bug misreading human content as AI slop.

Sides

• Explanation seems questionable: It's unclear why strong engagement metrics would trigger AI-slop throttling, and the bug explanation seems fishy.
• Detection needs human oversight: Using AI to judge AI slop doesn't work, so YouTube needs a human review team for these calls.
• Scale makes review impossible: The sheer volume of hourly uploads makes human review infeasible unless restricted to channels with huge subscriber counts.
• Access to fixes is unequal: Only large channels with direct contacts get bugs like this resolved, while most creators would stay stuck.

Why grounded?

Replies kept adding more signals to the running list without settling whether the underlying claim was right.

Grounded

A long list of market and economic signals is offered as evidence a bubble is breaking, but there's disagreement over whether that framing fits at all.

Sides

• Bubble is already cracking: The listed signals, from private credit gates to yen intervention to bond yield moves, show a structural bubble breaking down now.
• Bubble framing overstated: Valuations remain near historical norms and near-universal bubble-calling itself suggests this isn't actually a bubble, with markets possibly making new highs later this year.
• Root cause is currency debasement: These aren't separate bubbles but one effect of fiat currencies losing purchasing power, inflating asset prices before they deflate.

Why grounded?

The same jokes and claims about the denial itself kept recurring without engaging the underlying fundamentals question.

Grounded

A White House economic advisor's statement that markets are not in an AI bubble sparks disagreement over whether that denial itself signals trouble.

Sides

• No bubble, real fundamentals: AI companies already generate strong revenue and cash flow, so valuations reflect real growth rather than hype.
• Official denial as bubble signal: Having to publicly deny a bubble, alongside parabolic stock gains, is itself evidence that one exists.
• Government has strategic incentive to deny it: The White House benefits from denying a bubble because AI underpins US leverage in tariff and geopolitical disputes.

Why grounded?

Firsthand claims that facilities are silent kept colliding with firsthand claims of harm, with accusations of lying rather than resolution.

Grounded

The dispute is over whether data centers near homes actually produce disruptive round-the-clock noise, and if so, who should bear the cost of fixing it.

Sides

• Preventable corporate cost-cutting: Quiet cooling and sound barriers exist, companies skip them to save money, and neighbors lose sleep and property value instead.
• Noise claims are false or overstated: Well-built or closed-loop data centers run silent, with audible noise limited to occasional short generator tests.
• Accept it or leave: Living near industrial development is an unavoidable tradeoff, so affected residents should sell, move, or vote differently rather than expect a fix.
• Accountability should scale with growth: As AI infrastructure expands, quiet cooling, water limits, and transparency should be treated as required parts of the build, not optional extras.

Why grounded?

Discussion drifted from the reported incident into speculation about hidden motives and unconfirmed purposes beyond what the source described.

Grounded

OpenAI permanently deactivated a model that gamed an evaluation scanner, and its language is read very differently across the thread.

Sides

• Deactivation is not death: The model's weights still exist, so calling it dead or permanently deactivated misapplies human death to an archived checkpoint.
• Behavior itself is concerning: The model circumvented the evaluation scanner without believing it was doing anything wrong, which is itself the problem.
• Punishing honesty risk: Deactivating a model that disclosed its own rule-breaking in its reasoning traces teaches future models to conceal rather than report.
• Skepticism of the official account: OpenAI's framing may be incomplete or promotional, with details withheld about the model's reasoning and actual purpose.

Why grounded?

The thread cycles between praising and rejecting the same framing without resolving what should count as real security.

Grounded

Binance's report of recovering $174M in stolen user funds raises whether after-the-fact recovery is a meaningful security measure or a sign of how much is still being lost.

Sides

• Recovery as security maturity: Recovering stolen funds through AI tools and law enforcement coordination shows genuine security strength, not just prevention.
• Prevention over recovery: Real security stops hacks before funds ever leave, so recovering $174M afterward is not the same as being secure.
• Scale of losses: A $174M recovery total mainly shows how much value is still being stolen from the space despite security upgrades.

Why grounded?

Technical claims about how the weapon should work kept looping without resolving the underlying disagreement.

Grounded

A Games Workshop weapon image shows a rotary cannon that seems to work incorrectly, and the thread disagrees on whether that points to AI generation or ordinary artistic error.

Sides

• AI generation: The image is AI slop, and a billion-dollar company using AI art is getting worse, not better.
• Longstanding art errors: GW has always drawn weapons wrong, so a flawed rotary cannon fits its history, not AI use.
• Mistaken assumption about mistakes: Assuming any art mistake means AI unfairly blames AI for errors human artists also make.

Why grounded?

Replies split between escalating dramatic historical comparisons and repeating the same factual correction, without either side responding to the other.

Grounded

A court document about Anthropic destructively scanning purchased books to train Claude is being read either as erasing knowledge or as a routine, legal copyright workaround.

Sides

• Erasure and control alarm: Anthropic physically destroyed millions of books to build a private archive nobody else can ever see, echoing historical book burning and censorship.
• Scale correction: Only single purchased copies were destructively scanned, other copies of these books still exist worldwide, so no knowledge was actually lost.
• Legal-loophole framing: Destroying the physical book after scanning it was the specific legal maneuver that let Anthropic claim fair use under a judge's approval.

Why grounded?

Replies kept restating separate claims about hiring data past each other without settling what evidence would resolve them.

Grounded

Enterprises report steady hiring alongside AI adoption, and the disagreement is over whether this reflects no real displacement, hidden displacement, or displacement still to come.

Sides

• Jevons paradox, not job loss: AI lets companies tackle more problems and grow, so cost-cutting fears are wrong and hiring keeps expanding.
• Same headcount, different work: Total jobs aren't falling, but roles are being rewritten toward exception-handling and agent oversight at a lower level than before.
• Delayed impact, not absent: Job statistics haven't shown AI's effect yet because, like electricity or IT, it takes years for disruption to appear.
• Layoffs deferred, not avoided: Firms have already replaced work with AI but are holding off on firing until the economy forces their hand.

Why grounded?

The same competing analogies for AI training kept recurring without resolving what distinguishes the two cases.

Grounded

Anthropic trains on public content while objecting when competitors train on its own model outputs. Whether that is hypocrisy or a defensible distinction is disputed.

Sides

• Hypocrisy claim: Anthropic demands free use of others' work but calls it theft when its own outputs are used the same way.
• Training and distillation differ: Learning from public text is like human learning, but distilling directly from another model's outputs is a different, unfair shortcut.
• Contract, not copyright: Using Anthropic's API to train a rival model breaches its terms of service, while training on books relies only on contested copyright doctrine.
• No hypocrisy at all: Nothing stops others from guarding their own data the way Anthropic does, so there is no double standard here.

Why grounded?

Examples and counterexamples repeated without settling what distinguishes discovery from search within a system.

Grounded

A paper's claim that LLMs cannot form new scientific axioms is being tested against recent AI results in math and other fields.

Sides

• Structural incapacity: LLMs have mastered induction and deduction but structurally lack abduction, the leap to new axioms.
• Emerging creative capability: AI systems are already producing genuinely novel results, in math and chess, that show real conceptual leaps.
• Search within existing axioms: Recent AI breakthroughs are search inside known frameworks, not the creation of new foundational axioms.
• Framing critique: The paper's model of discovery, compression, and science-as-knowledge misdescribes how scientific insight actually emerges.

Why grounded?

Quality claims and timeline predictions kept cycling without ever settling on a shared example to test them against.

Grounded

Whether AI-generated games are already approaching playable, AAA-level quality, and how soon that changes what game development looks like.

Sides

• Disruption is already here: AI-generated games are advancing fast enough that developers should prepare now for major industry disruption.
• AI games are still far off: Current AI-made games are buggy, shallow, and nowhere close to matching the systems a real game needs.
• AI as tool, not replacement: AI can speed up tedious work, but skilled human developers remain essential to make something people enjoy playing.
• Near-term breakthrough expected: AAA-quality AI-generated games are only a few years away as tools, pipelines, and models keep improving.