AI is too competitive for anyone to make money at it. Last year OpenAI made a browswer to get ChatGPT traction, and Google “had to” do AI mode. “Making an extra $B is not cool, you know what is cool? $1T eyeball/query share”. There’s too much “hype metrics” focus to juice stock/IPO values, and AI bubble pop needs to shift from circular financing to outright cannibalizing each other.

Deepseek is an incredible lab, and Xiaomi espcially and several others from China, surprising. US models are good too, but the overspending has no competitive chance vs open models, or China’s hardware advantages and progression.

had founded a translation startup

Doesn’t Cisco know any networking and firmware experts, with extensive legacy code knowledge, who could vibe code through a well thought out plan?

Anything except banking

Quick add AI to your bittorrent clients, and take deposits for your crypto/gambling exchanges!

Hawaian pizza was “invented” in the 60s in London Ontario area.

It’s people like you that require state to have robodogs with machine gun turrets on them.

Much less debt and financial “engineering” in China, and much lower borrowing costs.

Most web sites have a more promotional attitude towards their product descriptions. I guess big Newton Pippin industrial complex lost out to more well funded nazi apple propaganda.

Coreweave makes long term revenue deals before building anything, but they charge too little. They are forced to accept whatever price hikes Nvidia tells them to pay, and are running out of cash in next 2 months, with borrowing rates over 15% to get more cash. In last quarter, their average interest rate was over 8%, and even if they used a 10 year depreciation schedule, they would still have lost money.

When hourly GPU rental prices go up, coreweave doesn’t make more money. They have tiny spot inventory. It is also a lie that there is gpu shortages, because if there were, there wouldn’t be such rapid model releases. NVidia GPUs are only competitive at training, but if there were any genuine scarcity, new models would come out slower so that inference/revenue making could be prioritized.

Definitely massive risk. They are giving away about $200-$400/week user credits for their Muse agent. Hardly anyone would need more than that, but it requires giving them credit card info. Massive competition being released in this personal agent space, just a few weeks after their rollout.

SpaceX is going even more nuts. Musk is a fraud supported by Zionist first political rule over US with support for his fraud, but 1.3gw of “spec” datacenters comming online in next 3 months is 5x the hourly gpu rental inventory in the US. BS goal about 10gw extra in 2027 is still gross overspending that makes everyone else go bankrupt from Nvidia price hikes. Grok and Cursor revenue combined still under $1b (last quarter). Zionazi circular financing for GOP corruption will starve rest of industry from access to capital.

IPOs are failing everywhere. UAE being forced geopolitically to bail out OpenAI with $30b, because they are out of cash. Coreweave needs $2b cash just to make it to end of year, but no one will give it to them. SpaceX in similar cash burn rate.

SpaceX and Meta are both building to just rent GPUs to desperate labs without firm contracts. All labs have separate plans. Coreweave gives massive discounts that force structural losses before impossible interest costs, and Nvidia’s power to charge whatever it wants for delivery. Collapse is very near.

delete x accounts.

They need to add braille to written drivers test. So prejudicial. The actual car test is easily solved with a fitty.

There is a genuine “training corpus” that is an accumulated asset and grows and feeds new models. The vast majority of training cost, that should be expensed, is the training time compute. Neither of these are expensed. There was about 13 major US model releases in September. At least 7 of these replaced a 1 month old previous model family. Open AI gpt sol went from 5.6 to 6 to 6.1 in about 2 months.

Fast model releases not only lock in structural losses, they juice revenue growth illusions. Model use rate generally drops by 66% 4 weeks after release, because of competition, and a surge of use on release for evaluation.

Key figures have yet to be leaked on S-1, which should be released fully soon. Structural losses and cash burn may be much higher than hopes for growth.

Musk giving 1 share (about $160) to all US children with a Trump account, will make the next generation cheer for skynet genocide and oppression. “For a while, it added 10-20% to my net worth.”

with deep accounting tricks.

Full S-1 on Q3 losses and training costs needs to get published. Quarterly revenue growth rate was half of Q2’s. Not good if it keeps halving, while their committed compute costs rising exponentially.

$8B real loss on $16B revenue. ($72B is a bondholder gain (accounting loss) from conversion). $9B in training costs not counted as loss, even though it is worthless whenever next model version is released.

$561B in forced spending on compute over next 10 years, with $750B total compute costs lined up. Structural losses are baked in for a few years.

Providers of voting machines have skewed GOP in their donations since Bush 43. Variations of voting results from polling have been strongest in Ohio and Florida.

Obviously this is yet more extreme corruption, but election rigging is possible/likely happening without this embarrassment. Vote rigging is more subtle than buffoon in chief thinks it should be.

Couldn’t spare half the squares on my board.

There are US hosted deepseek llm access points. openrouter, zen are local, but its more groups that can spy/train from you.

you don’t have control of the version of the model being run, the hidden system prompts, or any co-agents that might modify output before its sent to the user/service.

A bit of a paranoid concern. Your worry is that service may try to lie to you, perhaps because of your nationality. There have been quality issues from US served Chinese models. Completely free access points will get ultra compressed/quantized/lobotomized in order to be cheapest to serve but also give a poor impression of the model. Q4 quantization is considered a good balance between cheaper and full quality. But that will still make more mistakes, and US services can fake providing full precision. Every model can choose to lie to you, and you can screen shot it. Using a secret liar model in place of advertised model is possible from any provider, but they are incentivized to be as good as they can be.