sberens

For comparison, openrouter says opus 4.8 is ~55 tokens/s and fast mode is ~102.

750 tokens/s for their largest model is going to be nuts

great, just confirm you also have >>$200MM revenue[0] and have also previously founded a hard tech startup!

[0] https://www.theinformation.com/briefings/midjourney-revenue-...

Also fmr cofounder of leap motion, which developed a mouse that didn't you to touch it!

I don't understand how people can hate on this. It's probably the most novel & ambitious consumer health device ever? Plus they're doing it fully bootstrapped. Let them cook!

If all you need is the idea, then the minions could easily start their own company and reach a similar order of magnitude outcome.

Investors are usually prevented from selling until 180 days after IPO

They've got, ballpark, $5t to $10t to make back in the next 5 years

OpenAI's spending commitment is in the ~1T range for the next 5 years, and Anthropic is ~300B.

If they continue to show strong growth, they likely need to be at 100-300B in revenue/yr to support their yearly payments + financing, not 1T.

Have you seen https://fort.cx?

As someone who likes bright monitors, I'm excited to try the 2000 nit peak brightness! Are there any comparable monitors to the XDR brightness wise?

I did use it for this project! While the UI was good, there was a lot of last mile debugging to make charts/animations/responsiveness work well.

Glad you like it - we're working on improving the airflow right now actually! Thanks for the feedback on the paint.

Thanks so much for the kind words and being an early supporter <3

Because the lamp is 6'3" it's only below eye level for people who are 6'7-8.

We had another 6'5" customer who was worried about the height but they said it was totally fine even with shoes on.

We had to remake half the mold, and I split it 50/50 with the factory.

Just made one - use HN100: https://getbrighter.com/discount/HN100 :)

LEDs are pretty insane these days - the ones we use have an L90 (time until they hit 90% brightness) of >50,000 hours (17 years if you use it every day 8 hours a day).

We use constant current reduction dimming so there's zero flicker!

Thanks! The ups and downs of startups are very real (maybe doubly so for hardware)

It's 60k lumens now, and it draws 580W off the wall

Looking back I agree it was miraculous lol, I don't know if I'd do it again...

The main cost driver is the sheer size/weight/power. Dimmability, adjustable CCT, and smart home controls do add a decent chunk though.

Thanks :) It turns out "hardware is hard" isn't an exaggeration!

Yep lumens are additive (though your eyes perceive them logarithmically).

I don't know much about car headlights, but chatgpt says high beams are typically 25-45 watts, and assuming a generous 200lm/w that gives you 5000-9000 lm.

Roughly speaking, it's expensive because it's 50 lbs & tons of electrical components (that are much higher quality than $24 headlights).

The controls weren't working because we had wired them up according to the labels which were wrong (which is also why the measurements didn't make sense to us).

Author here, happy to answer any questions about the journey!

Coming soon! Some people have tweeted about us unprompted:

https://x.com/jaybaxter/status/2001729207873999094

https://x.com/FeifanZ/status/2001168758740803781

https://x.com/VivaLaPanda/status/2000758151851966770

https://x.com/__venki__/status/2000378422456148081

I'm working on shipping the second batch of the Brighter Lamp, a lamp that's equal to ~30 normal lamps for daylight levels of light indoors.

We delivered our first 500 units last month and got positive reviews, but lots of small issues to straighten out.

[0] https://getbrighter.com/

This Reuters article claims OpenAI is going to generate $3.6 billion in revenue this year, but the costs will lead to a loss of more than $5 billion. It expects a major revenue jump next year to $11.6 billion

The article linked[0] is from last year.

A recent article[1] from this year says "OpenAI looks to meet its full-year revenue target of $13 billion and a cash-burn target of $8.5 billion, the report added."

[0]https://www.reuters.com/technology/artificial-intelligence/o...

[1] https://www.reuters.com/technology/openais-first-half-revenu...

Is "parallel test time compute" available in claude code or the api? Or is it something they built internally for benchmark scores?

Interesting there doesn't seem to be benchmarking on codeforces