scosman

29 ms·

511.
by scosman 2y ago
> Native audio input including tone understanding I can't seem to find docs on this. Have a link?
512.
by scosman 2y ago
Probably fair that he led the move to enterprise. Also true it blew a very very strong lead in the B2C business.
513.
by scosman 2y ago
> There's way more money in Enterprise B2B than there is in B2C. Definitively not true by a huge margin.
514.
by scosman 2y ago
CEO of the PC company can't miss the new primary form of personal computing and blame a VP.
515.
by scosman 2y ago
Having worked on mobile there 2008-2011: no, I would say he’s properly rated. Had years head start on the Android model (hackable OS+vendors+wide HW support), and threw it away. Chased iPhone model briefly without the commitment to actually
516.
by scosman 2y ago
The sponsors aren't partners who are tuning up the bikes. It's companies who cut a cheque and expect logos everywhere. Individual riders/teams are motivated to win and get sponsor dollars, so may resort to cheating. They know
517.
by scosman 2y ago
> your best bet is to learn how to start and run a business Startups are so hit and miss. It's really really hard to make $500k a year at a startup. There's many years of zero, near zero or sub-intern salary. After that there&#
518.
by scosman 2y ago
They could donate to their dependencies that need it? Still directly supporting the code running in the project. Balances out that people seem happier to donate to the UI/product layer than something like ffmpeg. Might be a bad example
519.
by scosman 2y ago
There’s a decent amount of “good news overnight” because of time zones. Samsung release earnings with higher than expected phone demand, Apple will drop. Apple releases higher device projections, TMC will rise. Nokia releases new projection
520.
by scosman 2y ago
You can use HTML with plain text fallback. That's still the vast majority of marketing/form emails. HTML works fine, and adds important features like formatting text and links. HTML without images is ideal, as images are often blo
521.
by scosman 2y ago
They say no llama.cpp support yet, so no ollama yet (which uses llama.cpp)
522.
by scosman 2y ago
Did you try 1.5x generation speed and hear an improvement? Mine was totally just a guess. If it is better and you can beat traditional audio books on quality that would be cool. A lot of people listen at higher speeds. Re:subscription - may
523.
by scosman 2y ago
Great project. Pricing: maybe try a mobile app with monthly subscription? Something for recurring revenue. Features: can you generate at 1.5x speed? Might be more natural than the playback speed up options and be a nice differentiator.
524.
by scosman 2y ago
Every one I’ve used is on AWS and will tell you the zone so you can keep things within the region/zone. That includes Heroku, Supabase, Neon, and (obviously) RDS. That said, I’ve used AWS DBs with Cloudflare pages and the latency isn’t
525.
by scosman 2y ago
Codium also cool. For open source projects you can just ask it to review, no install needed.
526.
by scosman 2y ago
I’m reminded of what a beautiful language objective C is. It completely solves this problem. For example, take a look at this iOS method name (real Apple API): initwithenablefan:enableAirConditioner:enableClimateControl:enableAutoMode:airCi
527.
by scosman 2y ago
Okay, coderabbit.ai is pretty good! Nice UX. I recommend trying it out.
528.
by scosman 2y ago
Sync rights are expensive and complicated. Not covered in your Spotify subscription.
529.
Ask HN: Any great ML code reviewer tools?
4 points by scosman 2y ago|2 comments
530.
by scosman 2y ago
Re: 7b models vs gpt-3.5, I’m guessing different fine tuning parameters can account for the difference. The OpenAI fine tuning is a black box.
531.
by scosman 2y ago
That’s still the point. That model now does exactly one thing, and because of that can do better than a model 50x the size that tries to do everything. It will crush it in instruction following and consistency. A fine tuned 500b parameter m
532.
by scosman 2y ago
Wild to see them advertising collecting GPT4 responses for training other models. That’s definitely not allowed by TOS. I suspect many do, but front page advertising is another thing entirely.
533.
by scosman 2y ago
On that note: is there a good service for “here’s my dataset”, please fine tune these 9 models and give me evaluation stats?
534.
by scosman 2y ago
And that’s the point of fine tuning models. Still good to see someone walk through their fine tuning process, with a mix of hosted and local options.
535.
by scosman 2y ago
Fees are lower for the merchant. As a result it is less likely to see a 50c surcharge to use a card with debit. Places like convenience stores often only have debit. Personally I use credit for anything major, for the protections. People wh
536.
by scosman 2y ago
It is possible to learn from private data while preserving privacy. Intro here (there are many other forms): https://machinelearning.apple.com/research/learning-with-pri... Lots of issues come with privatized ML though
537.
by scosman 2y ago
MySQL is GPL not AGPL. Very different.
538.
by scosman 2y ago
Love this. Reminds me of the “flip flap” boards in old train stations. Like https://www.vestaboard.com/
539.
by scosman 2y ago
Fun concept. If the creators read this, I suggest some ways of building trust. There’s no “about us”, no GitHub link, etc. It’s a random webpage that wants my personal details, and sends me a “exe”. The overlap of people who understand what
540.
by scosman 2y ago
New vs used is mentioned, but kinda critical for the comparison. Yes a $60 board is definitely less capable than a $300 PC. The cost difference is primarily driven by the factors compared: better CPU, better IO, and more memory. You get wha

More ›