Where the AI market is talking

Ask HN: I spent personal money on game dev. How to turn stalled project around?

One of the reasons that has caused hesitation in deploying the idea early is the possibility that the mechanics might be instantly swallowed up by a publisher who could replicate it in a matter of a few months. However, if player's don't care about the gameplay, no publisher will either. So that's a very practical piece of advice in your comment. Building the audience to validate the idea should take a higher priority. On the positive side, the studio showed enthusiasm across the team while working on the project (biased as it may be)

Simulate cassette tape audio profiles using FFmpeg

I read that audio engineers actually use specialized and very expensive ripoff hardware in their signal processing chain to replicate the effect of the electromagnetics (hysteresis?) of tape systems on the audio. To me, this seems like wasted effort. Any such filter can be implemented using DSP.

Corporate America Has Suddenly Decided to Stop Blowing Money on AI

> Chinese models have government-enforced censorship, while American models have security and legal restrictions. They are both government enforced. The USA administration is using early access and export control restrictions to do their enforcing. As for the effects: while I'm told the Chinese government is sensitive to some internal political topics, I've never seen it myself. I guess that's because I don't write code dealing on Tiananmen square or similar topics, and I don't know anyone who does. In contrast the restrictions on the USA models are a right proper pain in the arse. Hugging Face being forced to swap Chinese models to defend against the OpenAI GPT attack is a good an illustration as any. That said, these USA restrictions are a new thing that seemed to start with Mythos -…

Show HN: Anyclaude-SDK – Claude Code-Style SDK for OpenAI/Anthropic Endpoints

Kimi K3 is not cheap

No I think you're right that the amount of compute spent on office work is lower than coding - although I don't have any sense for the right share. The best source I could find was an OpenAI report [1] which mentions that ~66% of enterprise token generation is via Codex, which I would expect to skew entirely towards coding. But it's hard to say how the remainder is split, what proportion is 'frontier', or whether it's representative for Anthropic. On your questions - I've spoken to a number of execs and seniors behind closed doors but nothing public I can point to. Anecdotally, I've spoken to senior leaders at banks spending billions of tokens on one-off tasks like prepping execs for earnings calls or piloting end-to-end agent workflows for specific use cases (but mostly piecemeal/on…

The relay market powering token resellers and fraud

About the same way you'd know Anthropic is getting you the model you picked. Which to me is not really obvious, given clear differences in quality throughout the day and month

Elevated Errors for Opus 5

Anthropic is not fully in control of the model.

Google Discloses $94.1B in SpaceX Stock, Marking 6% Stake

Just like OpenAI/Anthropic vs open models it's way too early to call winners or losers in the self-driving market. I called it a Tesla R&D project on purpose. They are at least in 2nd place to Waymo but it's barely a market in 2026, Waymo has been very conservative with expansion the last 5yrs for a good reason. It was still R&D. Ramping up Cybercab production is definitely high risk play. Which is counter to OP's point that Tesla is now just some boring non-visionary company, especially combined investments in humanoid robots and semiconductors. If Tesla was just looking to make safe easy money they wouldn't be making huge gambles on new markets.

Third Drone Shot Down in Three Days in Romanian Territory

Where are you from? You see, I've been in Ukraine in 2014, and I've been observing everything from the front seat, so to say. Generally what ruskies say does not matter. Not a single word. But specifically, the pretexts they may be stating now (I don't follow what they say), may be different from what they stated in 2014. It baffles me that after so much bs spewed by russian propaganda, people waste their effort trying to synthesize a coherent story. The only rock fact: ruskies say whatever gives them tactical advantage at the moment.

Anthropic versus the entire tech industry

Google Discloses $94.1B in SpaceX Stock, Marking 6% Stake

Google also has around a 14% stake in Anthropic (as at 2025 [1]) and they participated in the "G" funding round [2]. Amazon has a significant stake in them too. [1] https://www.nytimes.com/2025/03/11/technology/google-investm... , https://archive.md/f0BxC [2] https://www.anthropic.com/news/anthropic-raises-30-billion-s...

Kimi K3 is not cheap

But if you compare to Anthropic's models? The cost difference is huge. Anthropic is clearly concerned that people are realizing they are expensive, since the Opus 5 blog post dedicated a lot of time to talking about how cheap the model was compared to the competition... but this doesn't hold water when I haven't seen any independent benchmarks claiming Opus 5 is cheaper than GPT-5.6-Sol, even if it is supposedly closer. GPT-5.6-Sol is pretty competitively priced, but not all American frontier models are, and even 10% to 30% is still significant for any commodity that's as fungible as frontier models often are. > as you say that could go down to 20-30% cheaper I never said anything about 20% to 30%. We don't know how much it actually costs to host this model yet, and that will determine…

The relay market powering token resellers and fraud

Otoh I don't care about the ethics of doing this to big companies like OpenAI and anthropic and not small ones

Kimi K3 is not cheap

Agreed, Kimi is cheaper for coding - I say that explicitly in the post too. However I'd have to disagree with you on the "office task" front. General office work is one of the big frontiers the labs are pushing on, and it's part of how they're justifying the value proposition to enterprise customers. It's also accounts for a big portion of the spend on RL; tasks/environments designed to train agents to navigate Slack or Salesforce. If you're Anthropic pitching Claude to a bank (taking an example I'm familiar with), coding probably accounts for ~20% tops of the workforce, and it doesn't drive direct revenues. The 'agentic coding bump', but for all your analysts, traders, and wealth managers, would be a much more attractive prospect. I don't disagree that coding is the most successful …

Ask HN: What's the best hands-on path to learn ML inference infrastructure?

Since Anthropic trains on Sagemaker on AWS, what about some Cloud Certifications?

DeepSeek pause fundraise after comments on compute gap to US leaked (transcript) [pdf]

And that's why OpenAI bought all the future ram contracts

I learned PCB design, 3D printing and C just to listen to music

I have serious beef with ChatGPT's text-based circuit schematic drawing abilities. Even though the model itself appears to be fully coherent about the big and little picture aspects of whatever's "on bench" I have wasted too much time trying to parse its attempts to draw circuits. I consider them actively harmful in their current form. I quietly hope that one of the LLMs will start actually generating netcode that can be pasted into KiCAD, or even rendering circuit snippets inside of the chat stream properly. I've seen KiCAD running in the browser, so it's not like this isn't completely doable. It's just a question of resource allocation.

Kimi K3 is not cheap

It feels like this "Kimi is a token hog" meme is 100% astroturfed by Anthropic. It's cheap. Believe your own eyes.

OpenAI: A Bubble Bigger Than Dotcom

The new rules of context engineering for Claude 5 generation models

Yes, it IS confident of its weights & biases... but, if you keep at it, Claude WILL find the smoking gun, eventually. Even a broken watch is correct twice a day ( unless it's a digital watch, without a battery, in which case, it's just broken... ) But seriously -- newer Claude (and OpenAI and Google and ???) models DO find the smoking gun, if you let them keep going until they reveal the weird chain of events that leads to a bug. I was seeing the most obscure UART driver bug, where it would work at 1,500,000 baud (!) but fail by only outputting the 1st char at 230.4k and 460.8k -- and it was due to a very narrow race that would check the buffer, if not full, insert a character, and return BUT sometimes the TX Complete interrupt would happen between the check and the insert, and something …