• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

The anti-AI thread

Page 37 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Actually, from what I've read, it is coming out now that OpenAI doesn't have the money to pay for that remaining 40% of the DRAM market that they tried to scoop up. Adds up to like 900,000 wafers a month.

I guess the only question is that, if they can't buy it, will someone else? Or, is the AI pimple just going to burst?
 
Steltek said:
Actually, from what I've read, it is coming out now that OpenAI doesn't have the money to pay for that remaining 40% of the DRAM market that they tried to scoop up. Adds up to like 900,000 wafers a month.

I guess the only question is that, if they can't buy it, will someone else? Or, is the AI pimple just going to burst?
Click to expand...
The AI Jörmungandr will bust like fireworks and I will have some popcorn.
 
sdifox said:
ASML shipped FIVE HI-NA EUV scanners in 2025. Currently waiting time is 12-18 month for a EUV scanner.
Click to expand...

So you put in an order, and wait. Rome wasn't built in one day. A fab takes multiple years to build but it will never get built if you don't at least start the process.

Personally if I had billions to burn like the AI companies seem to have I'd be starting with slightly older tech that might be faster to get and also slightly easier to use and design for. Start with that. Start pumping out DDR4 and older flash chips that don't require the latest gen EUV tech as they are still very viable today and still in high demand. Eventually upgrade to EUV. The wait time is actually a blessing in disguise, since you still need to build the actual fab building, clean room, other supporting infrastructure, and also hire people that actually know how to use those machines. Most likely get training from ASML as well. I assume they provide that. Also need to procure all the raw materials and chemicals. All that process would take as long as it does to get the machine and actually have it installed, so once you do have a fully operational machine you already have staff that know how to use it and can now start to do test runs. That might take a year or two before usable chips can be produced. Ideally you want at least 2 production lines for redundancy as well but probably best to start with one, refine it and work out all the kinks, then build another.

It's a long process, but it eventually gets done. It just seems odd to me these companies are spending enough money to build multiple fabs on something that so far is not actually profitable, when they could simply use that money to build fabs. Even if AI completely dies out we will always need chips.
 
Kaido said:
Dang!! What are you going to use that for??
Click to expand...
1st is evaluating a vibe-coding setup disconnected from the cloud. Like how viable is it going to be evaluating a whole code base and doing multi file edits etc...
2nd is slipping lose a openclaw box and not worry about the credits cause it's local 😉.
3rd .. maybe dipping into the art of finetuning.

another motivator is just the factor of having capable local llms in terms of prepping. World going to shit? It might come in hand having a MacGyver 2.0 on steroids on hand.
 
Red Squirrel said:
So you put in an order, and wait. Rome wasn't built in one day. A fab takes multiple years to build but it will never get built if you don't at least start the process.

Personally if I had billions to burn like the AI companies seem to have I'd be starting with slightly older tech that might be faster to get and also slightly easier to use and design for. Start with that. Start pumping out DDR4 and older flash chips that don't require the latest gen EUV tech as they are still very viable today and still in high demand. Eventually upgrade to EUV. The wait time is actually a blessing in disguise, since you still need to build the actual fab building, clean room, other supporting infrastructure, and also hire people that actually know how to use those machines. Most likely get training from ASML as well. I assume they provide that. Also need to procure all the raw materials and chemicals. All that process would take as long as it does to get the machine and actually have it installed, so once you do have a fully operational machine you already have staff that know how to use it and can now start to do test runs. That might take a year or two before usable chips can be produced. Ideally you want at least 2 production lines for redundancy as well but probably best to start with one, refine it and work out all the kinks, then build another.

It's a long process, but it eventually gets done. It just seems odd to me these companies are spending enough money to build multiple fabs on something that so far is not actually profitable, when they could simply use that money to build fabs. Even if AI completely dies out we will always need chips.
Click to expand...
Because no one else thought of this before...

2026 TSMC capex is projected to be 50-60B. They only build when they have secured contracts for when the foundry will become operational.

Samsung is looking to spend 73B
 
Anthropic: "Put the genie back into the bottle!"

developers.slashdot.org

Anthropic Issues Copyright Takedown Requests To Remove 8,000+ Copies of Claude Code Source Code - Slashdot

Anthropic is using copyright takedown notices to try to contain an accidental leak of the underlying instructions for its Claude Code AI agent. According to the Wall Street Journal, "Anthropic representatives had used a copyright takedown request to force the removal of more than 8,000 copies...
developers.slashdot.org


nakedfrog said:
Silly sdifox, Squirrel is never wrong about anything.
Click to expand...
Funny that a MAGAt thinks that socialism is the solution to the AI bubble.
 
manly said:
Anthropic: "Put the genie back into the bottle!"

developers.slashdot.org

Anthropic Issues Copyright Takedown Requests To Remove 8,000+ Copies of Claude Code Source Code - Slashdot

Anthropic is using copyright takedown notices to try to contain an accidental leak of the underlying instructions for its Claude Code AI agent. According to the Wall Street Journal, "Anthropic representatives had used a copyright takedown request to force the removal of more than 8,000 copies...
developers.slashdot.org
Click to expand...

Hmm...

copyrightalliance.org

What to Know About the $1.5 Billion Bartz v. Anthropic Settlement | Copyright Alliance

At the end of August 2025, Anthropic agreed to a historic $1.5 billion settlement in the major class-action AI copyright lawsuit, Bartz v. Anthropic. The settlement brings the case to a close with Anthropic having
copyrightalliance.org copyrightalliance.org

1775162469208.png
 
Regarding the Anthropic lawsuit(s) over source code ... the magic is obviously not in their "source code" but their superior training techniques, it's the actual llm's not the source code that interfaces with them.
So no. Anthropic is not "naked" ... they got the same stack as everyone else, their models are just better.
 
Have they had an ai make itself yet? I wonder what the minimum would be to bootstrap it?

I ask, cause ai generated code isn't copyrightable. I wonder how that would work out in the courts, if a company's source code got released to the wild, but it was largely written by ai.
 
lxskllr said:
Have they had an ai make itself yet? I wonder what the minimum would be to bootstrap it?
Click to expand...

Yes: (well...assisted)

github.com

GitHub - askery/automl-list: A list of open-source and commercial Automated Machine Learning tools

A list of open-source and commercial Automated Machine Learning tools - askery/automl-list
github.com github.com

If you're in the OpenClaw ecosystem:

github.com

GitHub - aiming-lab/MetaClaw: 🦞 Just talk to your agent — it learns and EVOLVES 🧬.

🦞 Just talk to your agent — it learns and EVOLVES 🧬. - aiming-lab/MetaClaw
github.com github.com

Keep in mind:

1. All human endeavors begin with commitment

2. Development is iterative over time as (1) existing knowledge is gained, and (2) discoveries are made

3. Therefore, projects need guidance (goals) & progress over time (R&D)

This is the reason AI-building-AI is "assisted" & not "self-directed":

1. Humans have to mine the material, build the computer, supply the power, etc.

2. Humans have to define the problems to be solved. Ferraris & monster trucks are both "cars", but Ferraris are pretty terrible for offroad activities! We are now seeing niche-specific AI systems: Perplexity for news, Claude for coding, OpenClaw for execution, Ask Maps for live local travel, Nano Banana for images, Seedance for video, etc.

"The Matrix" & Skynet are temporary illusions due to entropy. "Infinite regress" is the defining factor here because eventually, you NEED a bootstrap system (i.e. human beings). WALL-E was a fun exploration of that:

1. Humans ruined earth
2. They built a self-sustaining ecosystem in space
3. Perhaps worse than the Matrix, they created Couch Potato Land™

But really, it would be more like the Hunger Games:

1. A rich Capital
2. Poor feeder colonies

The energy story alone is global:


The pipeline is vast:


For people worried about AI taking over the world...meh. The support system required for the current iteration of AI is ridiculously complex lol:

1775196058717.png
 
Despite being anti-AI personally, I can't think of any way that this translates through to being a good idea:
it.slashdot.org

Internet Bug Bounty Pauses Payouts, Citing 'Expanding Discovery' From AI-Assisted Research - Slashdot

The Internet Bug Bounty program "has been paused for new submissions," they announced last week. Running since 2012, the program is funded by "a number of leading software companies," reports InfoWorld, "and has awarded more than $1.5m to researchers who have reported bugs " Up to now, 80% of...
it.slashdot.org

If AI is being used to help determine verifiable facts, I can't think of a by-principle issue with it. Artistic expression is a different story of course.
 
mikeymikec said:
Despite being anti-AI personally, I can't think of any way that this translates through to being a good idea:
it.slashdot.org

Internet Bug Bounty Pauses Payouts, Citing 'Expanding Discovery' From AI-Assisted Research - Slashdot

The Internet Bug Bounty program "has been paused for new submissions," they announced last week. Running since 2012, the program is funded by "a number of leading software companies," reports InfoWorld, "and has awarded more than $1.5m to researchers who have reported bugs " Up to now, 80% of...
it.slashdot.org

If AI is being used to help determine verifiable facts, I can't think of a by-principle issue with it. Artistic expression is a different story of course.
Click to expand...

From what I understand, a lot of the so called "bugs" being found and reported using AI aren't actually bugs at all but rather reflect AI's inability to truly understand the code it is reviewing. Problem is, these so called "bugs" are then being dumped en-mass on the various projects as an attempted money grab with no prior human review and the people that have to verify them are being overwhelmed. This is precisely why curl ended their bug bounty program.


I fear that once the money leaves the system, the bug reports are going to dry up and we'll all be in a worse place than before.

And, at some point, AI is going to have to pay for itself. Providing hundreds of dollars worth of AI compute for $20 per month (like Anthropic, for example, does with its lowest tier paid plan) isn't financially viable in any sort of term.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top