ChatGPT serves ads. Here's the full attribution loop

https://www.buchodi.com/how-chatgpt-serves-ads-heres-the-full-attribution-loop/

99•lmbbuchodi•1h ago

Comments

gxs•1h ago

This is gross

It feels like we’ve been in the golden age and the window is coming to a close

Let the enshitification begin, I guess

2ndorderthought•1h ago

In the past month local models have been ramping up in major way meanwhile the namesake providers have upped prices, went offline randomly, and started doing slimier and slimier things.

I really think the future is local compute. Or at least self hosted models.

SchemaLoad•58m ago

The hosted ones still have the advantage of being able to search the internet for live info rather than being limited to a knowledge cut off date.

darepublic•57m ago

Local ones that support tool use can do the same

gbear605•57m ago

I’m not sure why a model needs to be hosted in order to make network calls?

hansvm•55m ago

Is there a library of good tools for LLMs to call? I have to imagine the bot-detection avoidance mechanisms are a major engineering effort and not likely to work out of the box with a simple harness and random local LLM.

ossa-ma•47m ago

Even the hosted ones are blocked from searching certain sites, for example Claude is banned from searching Reddit:

`Error: "The following domains are not accessible to our user agent: ['reddit.com']."`

wyre•42m ago

Tavily, Exa, Firecrawl, Perplexity, and Linkup are all tools for agents to search the web.

I’ve been building a harness the past few months and supports them all out of the box with an API key.

goosejuice•5m ago

Kagi also has an API. People who hate ads are probably the same cohort that should be paying for Kagi. That's the sane alternative world where companies respect their users.

eightysixfour•57m ago

You can do that locally too!

CSMastermind•57m ago

What's the rough equivalent of a local model? Are we talking GPT-4?

Terretta•51m ago

Depends on your VRAM or "unified" memory for how smart it is, and CPU/GPU for how quick it is.

128GB of RAM? Sure, the early to mid 4s releases, except maybe 4o. And on an M5 Max, about the same speed.

I wouldn't really bother under 64GB (meaning 32GB or less) except for entertainment value (chats, summaries, tasky read-only agent things).

kay_o•50m ago

GLM 5.1 and DeepSeek 4 are acceptable, but the cost of hardware and energy cost that depending on your use case you may as well purchase a Tokens. They get useless and stupid rapidilty if you quant enough to run on single 16-24GB GPU style.

2ndorderthought•32m ago

Qwen 3.6 which was released this month is a large but still smaller model. Supposedly it's at about sonnet level when configured correctly. It can be run on commodity hardware without purchasing a data center. https://www.reddit.com/r/LocalLLaMA/comments/1so1533/qwen36_...

Then there are middle size ones which require multiple gpus which are like gpts latest flagships.

Then there is kimi 2.6 which is a monster that is beating opus in some benchmarks. https://www.reddit.com/r/LocalLLaMA/comments/1sr8p49/kimi_k2...

It's basically whatever you can afford. Any trash heap laptop can run code auto complete models locally no problem. The rest require some level of investment, an idle gaming pc, or a serious investment

rnxrx•54m ago

The arc of the technological universe is short, but it bends toward enshitification.

dannyw•45m ago

How do you expect the spend & COGS for free LLM inference to be funded? For users who don't want to pay, or maybe can't pay?

infinite_spin•34m ago

From things like defense/private contracts

e.g. colleges pay for institutional subscriptions

2ndorderthought•26m ago

The average person doesn't benefit from defense contracts ... Like ever.

derektank•27m ago

Perhaps it’s a glib and easy thing to say, but after a teaser period, I would simply not offer free LLM inference. Agreeing to serve ads just completely re-aligns your interests away from providing the best possible user experience to something else entirely.

iammrpayments•34m ago

It has begun ever since they nerfed chatgpt4 before releasing 4o

singingtoday•1h ago

I don't like anything about this.

djmips•1h ago

And it begins.

WD-42•57m ago

Since they are served as distinct events then I would think they should be easy to block.

Once the ads are injected directly into the main response is when things get interesting.

lmbbuchodi•47m ago

you can block these URLs: |bzrcdn.openai.com^, ||bzr.openai.com^ It won't blanket block everything but will significantly reduce telemetry collected.

kardos•6m ago

> Once the ads are injected directly into the main response is when things get interesting.

This would be where you post-process the LLM response with a second LLM to remove the ad..

keyle•57m ago

Can't wait for "watch this ad for 90s to use xxhigh on your next prompt!"

uriahlight•49m ago

Let the enshittification commence!

vicchenai•45m ago

figured this was inevitable once they started the free tier. the attribution loop being a separate event stream is actually kind of clever engineering though -- means they can A/B test ad formats without touching the core model response

avaer•38m ago

Remember that ads are the "last resort" for OpenAI, and they're doing this despite the fact that it's "uniquely unsettling", according to Sam.

Was he lying, or has OpenAI given up hope that this train wreck works economically without enshittification? Neither option is good, but I don't really see a third.

Aurornis•19m ago

The ads are only for the free and $8/month plans. They basically added an ad-supported super discount level that you can ignore if you’re paying for the normal plans.

infinite_spin•38m ago

I see OpenAI making a significantly larger amount from defense contracts than from advertisements pumped into chats. So I wonder whose bright idea it was to create a public perception risk.

peddling-brink•32m ago

Maybe the negative press from ads is better than the negative press from powering murderbots?

tayo42•17m ago

Bad press from a contract like that happens once and everyone forgets. Ads are in your face everytime

Larrikin•27m ago

Every single MBA can show for at least one quarter revenue is up after they introduced ads. They do not care what happens after if they can plan their career around that.

jesse_dot_id•34m ago

That's cool, I'll never see them.

benleejamin•33m ago

I'd always thought that ChatGPT ads would be indistinguishable from actual content.

irjustin•21m ago

this would be a breach of trust and short term would work great but long term is too detrimental.

same thing could've been said for search results, so at least that part is still "safe".

bix6•19m ago

O you think trust matters? This is capitalism not trustism.

BoredPositron•26m ago

I don't get what's wrong with charging for your product. Like get rid of the free tier and make a small tier with an easy to serve model for like 5 bucks. Is it still the DAU rage of the 2010ss that's driving burning money?

teaearlgraycold•19m ago

How do you pick up new paying users without letting people use the service for free for a while first? Freemium is popular because it works well.

Aurornis•20m ago

The ads are in the free tier and the new ad-supported $8/month plan.

Every time this comes up there are comments assuming that ads are being injected into the normal plans, but these are for the free tier and the new Go plan which warns you that it includes ads when you sign up.

darepublic•13m ago

Would require a lot of training to implement ads blended into convo and not have it be too obvious/ eff up the results?

dankwizard•18m ago

Really well written, technical post. Good read.

mock-possum•9m ago

Not to me they don’t, cause I canceled my account and stopped using their products when they made the announcement.

Synthetic.new – Limits and Pricing

Show HN: 20-endpoint image API (remove bg, upscale, face restore)

A Git merge with 100k parents

How dating app algorithms (likely) work in 2026

Higher temperatures spur Alaska's invasive pike to eat more, bad sign for salmon

Epigenetic fingerprints link early-onset colon&rectal cancer pesticide exposure

An Explicit Solution to Black-Scholes Implied Volatility

OpenAI Wants Codex to Shut Up About Goblins

Chinese SUVs at Beijing Auto Show [video]

Fuck Off AI Music

The New Teams CLI

From One AI to Any AI: JetBrains rethinks the approach to AI tooling [video]

We decreased our LLM costs with Opus

Agent, Know Thyself (and bid accordingly)

AI Killed the MVP. What's Next?

Losing My Friend over Wegovy

Maladaptive Frugality

Going Full Time on Open Source

NBA proposes new '3-2-1' draft lottery system

Continuing the Story of Early DOS Development – Microsoft Open Source Blog

The Work Between Factories

Ask HN: What product analytics are you using?

Framework 16 Gets Nvidia RTX 5070 12 GB Upgrade Module for Eyewatering Price

Tencent used Anthropic's Claude to fine-tune it's new Hy3 AI model

CATL says sodium batteries are mainstream-ready, signs 60 GWh deal

Fedora 44

Amazon to offer OpenAI models on AWS after Microsoft exclusivity ends

Anti-Trump Instagram pic of seashells now enough to indict ex-FBI directors

Five takeaways from the King's historic address to Congress

Guinea Worm Disease