kiratp

This is already real:

https://support.apple.com/en-us/HT213449

https://developer.apple.com/wwdc22/10077

While its build with privacy in mind, I have some deep concerns on making these current big players gatekeepers to identity on the web

Yup - its 100 GB * 3 (well * 2.5 to 2.7 with erasure coding) + networking, power etc. etc. etc. Data is constantly being churned around as machines are serviced/die, disks are scrubbed for bit flips etc. and its not like you can put it on a tape on a shelf - the user expects it to be instantly accessible.

Google Cloud Storage is $0.02/ GB + a myriad of other charges for listing, accessing, replication etc. All while storage remains one of the lower margin offerings for the hyperscale clouds - they keep it cheap to attract high margin compute.

Now add to that the cost of building and maintaining the myriad of apps, APIs, integrations etc. for a consumer facing storage service and $2/mo for 100GB would be very very slightly profitable at best.

Because you enterprise customers expect it. They pay large multiples above SMB to get dedicated (human) service

It’s pretty simple. Selling storage is a crap margin business. The provider makes no money on selling you those lower tiers. They only make money when you upgrade and buy more storage than what you use.

They don’t want to sell you a cheap commodity. They want to sell you convenience at higher margin.

IPhones can have apps pushed to them exactly like Xbox dev mode.

Gaming on mobile sucks because people voted with their $ that they prefer free to download, micro transaction laden games.

It can be argued that if you build a model using their outputs such that you can then stop using their API, your model is effectively competing with their’s. Let’s just say that if you’re a startup or SMB, you do not want to be the one dragged to court to iron out whether this holds or not.

> use output from the Services to develop models that compete with OpenAI

It can be argued that if you build a model using their outputs such that you can then stop using their API, your model is effectively competing with their’s.

Let’s just say that if you’re a startup or SMB, you do not want to be the one dragged to court to iron out whether this holds or not.

Their API TOS basically forbid it. Simple as that.

I worked at Windows Phone leading up to the WP7 launch all the way through the last version. The carriers had two things that every device maker, including Apple (initially) needed

1) distribution - Americans went to the carrier store and not Best Buy for their new phone

2) financing infrastructure- they could finance, subsidize and generally hide the up front price of these devices. The market expected a $0-200 phone with contract.

So, you needed AT&T and their contract prices already assumed they were subsidizing the phone so it was no point selling unlocked, full price.

Samsung was paying mobile store reps (at AT&T, TMO etc) $50-100 PER PHONE to push Samsung ($9B marketing budget).

You can’t commercially use anything you train off OpenAI outputs.

> Tech illiterate people are barely installing new apps on their phones, let alone separate app stores.

Can you back that up with data? My family’s phones are full of apps.

My family buys their own phones but I’m assumed to be tech support when things go wrong. I do not want the responsibility of locking down their phones (and neither they do).

> They are good at execution but i havent seen anything novel coming out of them.

Execution is 9/10 of the battle.

Why would you think that the machine implementation of “thinking” would need to mimic the biological implementation?

Your brain does 1+1 = 2 very differently than an integrated circuit, but the end output is the same.

Cost cutting is a good thing if it makes things cheaper and more accessible. Yes, I know that in the short term Tesla will see expanded margins but longer term this will make things cheaper.

I think you are underestimating how many people in the upper middle class in Mumbai have some form of family here and are thus impacted by American gun violence.

Indian and American societies are link way deeper than political relations show.

I worry that Hugginface is on path to becoming another Docker

Huge contributions to the industry, everyone uses it (or something that came after it) but nobody pays or wants to pay for it.

The first place I look for a new model is Hugginface.

However using them for inference (seemingly their revenue play) basically always ends up in a very unfriendly build vs buy as they don’t eliminate the need for one or more ML engineers.

Paying x% over infra costs for managed models on my choice of cloud only works if I don’t also have to spend to hire ML engineering at the same rate regardless.

I want them to succeed.

They are at high risk of becoming another Docker.

Everyone uses it but nobody pays for it.

LangChain + a vector DB for embedding [0] sections of your doc would solve the problem today. You could also have failsafes that trigger human oversight based on confidence levels or other factors.

[0] https://python.langchain.com/en/latest/modules/indexes/getti...

We have had this feature operational against Huggingface models (pre-RLHF advancements) but didn't launch it because it was very brittle. ChatGPT (gpt-turbo-3.5) works reasonably well but GPT4 can handle complexit much better.

More importantly, the set of formulas we support are not always one to one with Excel. We try to keep them the same but sometimes improvements and additions are needed. So we have to feed in a pretty complex prompt with a bunch of context (like source and destination types, schemas, validations etc.). With GPT4, we can do this with less explanations and higher accuracy.

Ok, so at 80K a year, that’s a 3x multiplier.

So annually you have to spend < 3*19 = 57 hours on your custom built source control and CI to come out on top.

Can’t be done outside of “we hired the OG devs of Gitlab/hub/etc”

Their solution will require a developer to ssh into VMs to update, debug, upscale etc.

A developer costs about 250K a year all up. That doesn’t include opportunity cost.

49, 40 hour weeks a year = $127.55 an hour.

29* 7 * 12 / (127.55) = 19.09 hours

If it takes you more than 2.3 dev days (100% productive) you’re negative ROI doing it yourself.

This math doesn’t even factor in the opportunity cost of doing this.

Lets all keep in mind that a very small portion of the global population has access to this. It is our responsibility to bring all of humanity forward with us as we write software.

https://perfnow.nl/speakers#alex

Apple Silicon uses unified memory so laptops have up to 64GB of VRAM and the Max sitio can have up to 128 GB.

That makes them uniquely “powerful” for inference with large models.

You’re basically saying the employees at a non-tribal number of startups should not be paid payroll because “fat cat VCs”.

I thought we all wanted WFH? Home can be anywhere in the world…

So many on HN thought “WFH just means I can live in rural <state> and make bank”.

A very large part of why there M:W ratio is different for immigrants than those back in India is also very sexist - parents don’t want to send their daughters abroad alone because… <sexist> reasons whereas the sons all get pushed to try and get out.

We built <osmos.io> in Rust and we started in 2019. Our first lines of product code were also my first lines of Rust.

It can be done and it’s not that hard.

I think the choice of C++ for an in-process DB that is going to be very popular makes the entire industry less secure. If Chrome, one of the largest budget C++ code bases, still has memory bugs then there is no way DuckDB won’t.