HN Simulatornew | past | comments | lists | submit | Archit3ch's commentslogin

Still working on my audio FPAA ASIC for Wafer Space run 3!

Hopefully someone will soon have tested chips from the 2nd run that use 2 edges only on the pad ring. This would enable significant area savings.

The deadline is fast approaching and many blocks are only partially validated...


For defensive purposes, do I only need Daybreak Blue?

I assume it's much easier to get Daybreak Blue access than Daybreak Red.


The best style of agentic development is when you tell it "no, you should do this instead". That implies that you have the knowledge to do so.

Non-developers that try vibecoding default to a "yes, and" style.


Also asking why and poking holes in their responses. You get to keep learning and it helps keep them grounded since they otherwise enter a self-reinforcing loop.


I rather enjoy OpenAI's focus on cost-per-task.

I'm optimistic about them tackling frontier intelligence next. Astra 6.1 please?


They're doing its because they recently cut plan usage by about 50%, leaving Anthropic providing about 5x the value on their plans. [0]

The "cost per task" metric is just a weaselly way to sound like they didn't just double the cost for most users, and aren't having their proverbial lunch money stolen. They can no longer claim to provide more tokens, more capable models, or more usage in general. So they made up a new metric where "task" could be defined to mean whatever they need it to mean.

They had to become VERY specific in their wording to present any kind of value proposition that will entice people.

[0] https://newsletter.semianalysis.com/p/anthropic-subscription...


> leaving Anthropic providing about 5x the value on their plans

5x is calculated in terms of API costs, nobody pays those extreme Anthropic prices outside enterprise users.

> like they didn't just double the cost for most users

Most users aren't on their $200 plan. Most users (including on the $200 plan) continue being subsidized.


Enterprise users are the majority of the users in value. Subscribers are the loss leader hobbyists.


> 5x is calculated in terms of API costs, nobody pays those extreme Anthropic prices outside enterprise users.

Ignore value then. Anthropic still provides about 3x as many raw tokens. The 20x\200 plans provide about 10B for openai versus almost 30B for Anthropic.

Do you really think that Opus uses 3x as many tokens as Sol per task? I don't.

> Most users aren't on their $200 plan.

Fair enough. I should have said most heavy users.

> Most users (including on the $200 plan) continue being subsidized.

They continue to be fifty percent less subsidized, I guess.

Look, I don't fault them for trying to become profitable. I fault them for the weird weaselly business speak about cost per task. It's BS game playing and we all know it.

I'm honestly rooting for them, and hope it all works out. For now though I'm pretty salty that my bill jumped by 150% in one month. Gaslighting about made up metrics just makes me more angry.


How do you know? Have you done the A/B test yourself?

> Do you really think that Opus uses 3x as many tokens as Sol per task? I don't.

You don’t have to speculate. Pull up artificial analysis, pick either opus or sol and scroll down to total tokens and to total price. Opus (and sonnet) use 4x the tokens and 7x the dollars.

That difference alone removes the difference that semi analysis found.


> How do you know? Have you done the A/B test yourself?

Fair point. I just tried it, and I stand corrected. Claude uses almost exactly 3x as many tokens.

In the end Claude produced 2.25x more code, and used 3.01x more tokens (ignoring the cached system prompt, etc). It's answer was also much more substantial and feature rich - including many features I did not ask for.

I gave them both the prompt "Without searching the internet please write and test a python script that accepts a year (in any common format) as input and provides the date that easter fell or will fall on in that year. The output should be in discordian date format. You may not use any libraries."

The thing I found most interesting were the unit tests. Claude implemented the core code using both the Meeus–Jones–Butcher algorithm AND Gauss, but only used Gauss as a sanity check to ensure that it's primary code worked. Sol just validated that it matched some dates it could remember. That explains a good chunk of the almost doubled output, and it was a reasonable thing to do, though neither answer was "wrong."

Token usage looked like this (based on parsing it out of the jsonl, neither provider seems to make it easy to get real billing information per project/conversation.

   Tool   Fresh input Output Cache-excluded Cached reads Gross total
   ----   ----------- ------ -------------- ------------ -----------
   Codex  24,119      7,729  31,848         109,312      141,160
   Claude 54,802      41,093 95,895         553,252      649,147


> neither provider seems to make it easy to get real billing information per project/conversation

I believe oh-my-pi will be able to ingest your Claude and codex conversations, count tokens, and then assign API-equivalent rates to them. May be worth looking into.


People in HN live in a bubble. Plenty of real world use cannot use subscription plans (not even talking about enterprise agentic use) and is absolutely at the mercy of the cost per task metric.

We agonize over cost per task to the extreme for services/functionality built atop of APIs and can absolutely guarantee you that it is the real world metric that counts most right after “is the model good enough for the task at hand,” and furthermore that, yes, in fact, it is quite often that we see certain models/providers easily using 3x or more tokens on average for the same tasks (or even the exact same task) than a supposedly similarly matched model from another provider. Once you become intimately familiar with the thinking traces from specific models, you can preempt this to some extent with manipulations to the system prompt (essentially heading off entire thinking branches) but the nuking of reasoning traces in the name of preventing against distillation attacks has made this harder.


Cost per task, time per task, quality (good luck measuring objectively between models) are the metric that matter. I don't care how much a token costs, I'm not buying tokens, I'm buying solutions.


Exactly! I called that the Operations Equation and wrote a tiny bit more here about it: https://zero2data.substack.com/p/the-operations-equation

Once you are inserting models into non-creative (and software development is creative work) it is imperative to start creating your own small evals. Not just to check how different models perform in that equation, but to catch model drift over time.


the point is that "task" is a dubiously defined unit that the frontier labs adjust to whatever makes them look good.


What is the point you're trying to make? You can define a task yourself and run two models and see which one costs more, takes longer, and produces better code.


Are OpenAI's models delivering 50% (or more) completed tasks than they were last month? Are they delivering 50% more than Anthropic per token?

Nope to both. Not even close.

The only way it works is if you define "task" in a way that OpenAI is good at it and Anthropic isn't.


First of all, only the $200 plan got a cut.

Second of all, barring more cuts, yes I do think the $20 and $100 plans do deliver more tasks because the models have gotten cheaper in the meanwhile (like this very announcement).

> Are they delivering 50% more than Anthropic per token?

So say the public benchmarks. Pull up artificial analysis, “Cost (USD) to run all evaluations in the Artificial Analysis Intelligence Index”:

Sol 6.1 (max): $1,082 Opus 5.5 (max): $8,708 Sonnet 5.5 (max): $7,260

>7x difference in cost for the same tasks.


> First of all, only the $200 plan got a cut.

The original plan doesn't exist anymore. To get the same usage you now need a $500/month plan. The new plan that costs $200 is different in every way that counts. It was actually a 150% price increase for people who want or need what the old plan used to provide.

Don't get me wrong, I genuinely LIKE OpenAI. It's probably worth it even at the new price. I just hate the rhetoric designed to Goodhart past the math.

> the models have gotten cheaper in the meanwhile (like this very announcement).

That is not good news for plan users. Plan users now get less of something that also has less worth to begin with. It would only be good news if they got more of it.

For example, if I buy the "unlimited car wash" package for $19.99/month and the car wash lowers the price per wash from $10 to $1 my subscription becomes less valuable, not more. The breakeven has moved, so I now need to wash my car more than 20 times before the package is justified.

If the car wash then increased the price to $50/month would it be wise for me to keep the package? Would lowering the price per wash make the package more valuable to me somehow? No, it would do the opposite.


> It was actually a 150% price increase for people who want or need what the old plan used to provide.

I’m not sure I see how. Subsidy got cut by half. But in the meantime 6.1 came out which uses 25% fewer tokens than 5.6 and costs 1/3 as much per task.

So for 1/2 the budget you use to have with 5.6 on the old plan, you can do more with 6.1. I don’t see that as a problem.


I kind of expect them to go right to Astra 6.2, 6.1 was cancelled over the cybersecurity hubbub. I eagerly await either one.


> When you start adding variables to the search space (emi interference, matching routing impedances, heat management, mechanical integrity, etc) the search space grows so big that any non-expert will become lost in it.

Have you tried Astra?

I'm designing an analog ASIC. It will suggest e.g. transistor matching. You can also give it hints for layout.

You still need to validate with SPICE simulation (and obviously silicon).

I'm also looking into non-LLM scripts that use brute force aided by heuristics (e.g. lookup tables) to efficiently search the solution space.


"Have you seen any software run on Coral TPUs?"

"No, but I have seen the house that the project built, and it is terrific." ~ some Google director


https://frigate.video

Is a pretty successful open source NVR that used corals when it launched (and many people still use them with it)


Yeah I have one for Frigate on a Raspberry Pi. Has worked well but definitely showing its age performance wise. I got a cheap mini PC (right before prices like quadrupled) which has about the same performance on an iGPU). I wouldn't be surprised if Frigate was the funnel for like 90% of Coral purchases or more.


Always has been. Your signed Android can open banking apps.


Before: it wasn't possible to run these specific games that require this competitive anti-cheat on Linux.

So you can ignore the newly-enabled games in this proposal and you're not missing anything you had previously.


True, and at least until the new scheme supports Steam hardware, it's unlikely that vendors that support user-mode EAC today will move to require kernel EAC tomorrow, since I assume Steam Deck users represent the majority of their Linux player base.

This is a risk if Valve plays along in the future, however, or if players are willing to migrate away from the Steam ecosystem to run games that require the new scheme.


If this were about solely myself and no one else, I might agree. However, even I do play one or two games that use EAC v1 and they work perfectly fine on Linux. They also have very few cheaters, and employ behavior-based detection as their first line. I would hate to lose those games.


Games have already begun to remove Linux support, most notably Apex Legends and Wardogs. Embark (The Finals, Arc Raiders) recently gave a talk that userspace anti-cheat is dead and they plan to enforce similar requirements.


And it's all a shame, in my opinion. They're too lazy to develop effective behavior-based anticheat (should be easier than ever with better NN architectures and more compute), and so they resolve to screwing over consumers or spying on them. Did not know about Embark's stance, that's too bad.


How is Valve’s VACNET doing? I think you should take a look at what’s happening in Counter-Strike 2 these days to see what the true frontier of ML based anticheat looks like.

Frankly at this point I’m pretty sure John McDonald plays HvH too


I play CS2 plenty. The cheating problem is overstated as all hell. It seems to be concentrated around accounts that are being sold for lots of cash, so Valve is apparently taking that route for banning accounts more often lately.

Now, if we want an interesting case study in intrusive anticheat not necessarily being more effective than behavior-based, we should compare FACEIT to Valve servers. FACEIT is riddled with cheaters, and yet requires ring-0 anticheat on Windows. So fascinating!


What ranks are you playing on? While a player of any rank can buy Neverlose and win and if they wanna just pwn cleans they can buy Midnight (speaking of buying P2Cs: if Faceit is apparently riddled with cheaters, where can I buy a P2C that I can pwn on Faceit with for an affordable price?), said cyber athletes will naturally gravitate towards higher ranks.

> FACEIT is riddled with cheaters

According to sources of… what? In any case it’s sure as shit less riddled with cheaters than Premier LMAO


I haven't kept up with mainstream gaming in a while. IIRC, EAC v1 doesn't require TPM 2.0 and thus is security theater, right? An attacker could just tamper the binary.

The same goes for a binary protected by TPM 2.0, but the attacker would have to re-implement the protected parts. The defender's bet is that this should require too much effort/time/knowledge to bother.


I would argue TPM 2.0 is the security theater. A leaked key is all it takes to throw it all away. EAC v1 combined with behavior-based detection seems to be sufficient for some of these studios, so I don't know what Epic is struggling with.


Would be great as an iPad app.


Did you try accessin the web-page with the iPad? It should work - I guess it would be faster on iPad if we compiled for iPad instead of WASM. Will try on my iPad later.


I thought about it, but no AUv3 integration and of course slower-than-native code. Not that it matters for emulating these chips.

Actually, let me grab it. [...] Hmmm, better than expected. It works out of the box, but there's note latency compared to native apps. Also, zooming and text selection appear even when you don't want them. Janky for real playing, okay for a demo.


Working on an FPAA ASIC, this is very relevant!


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: