HN user
matthewrajcok [ AT ] gmail [ DOT ] com
Yes - I'm able to run Llama 3.1 405B on 3x A6000 + 3x 4090.
Will have Llama 4 Maverick running in 4bit quantization (typically results in only minor quality degradation) once llama.cpp support is merged.
Total hardware cost well under $50,000.
The 2T Behemoth model is tougher, but enough Blackwell 6000 Pro cards (16) should be able to run it for under $200k.
I've been on the receiving end of this - as an individual maybe, but as a committed startup not necessarily. Rooting for them!
Mercury has been great! Used SVB at my last company and it's a big improvement.
Agreed. But perhaps it could be bundled with something else, like working remotely as a junior engineer at a startup for min wage < x < SF CS grad comp.
This small risk of something very bad sounds like a case where insurance against this could be a win-win. "Pay a % of your increased earnings, and we cover any litigation related to your non-compete." If someone were able to analyze agreements and state laws to develop a good underwriting model there could be potential for an insurance tech startup here.
I negotiated quarterly billing instead of annual and a moderate discount for their cheapest enterprise plan, so there seems to be some room.
You need to give specific permission to the Mystro accessibility service for it to control the rideshare apps.
Thanks!
He was a driver full-time for 2 years.
When the driver is not on a trip, we turn on both apps. When they are on a trip, we turn off the app not in use.
We have a bunch of the top influencers in the on-demand space helping promote.
We've thought about pitching in the car but the cost doesn't work out.
We don't actually use their API.
It's much harder to do UI automation on iOS -- it is more locked down.
We have a pretty robust tool to keep our integration working with any app updates.
Thanks!
Definitely -- right now Android Auto is something we're very interested in integrating with.
Thanks! I agree with this, and we've actually seen many users buy Android devices just to use our app.
We actually are solving a massive driver retention problem for them (96% churn) so we see ourselves as beneficial to the rideshare companies since we increase driver earnings and retention at no cost to them.
We have a framework built out that lets us rapidly respond to app updates and a team of testers all over the US.
We have also thought about how to best circumvent any active attempts to block us -- since we interface with the driver apps directly rather than any APIs that would be harder to do.
Yes -- I think we're going to adjust this and also try to show them how much extra money Mystro is making them.
Thanks! All drivers can use for free with up to 10 trips a week, and there's a 7-day money-back window for the unlimited plan.
We have a good understanding of what will provide the best driver experience based on a huge amount of user feedback.
There is also the potential to use the data we have to optimize driver revenue even further.
The IRS will treat it as an option rather than a purchase if the shares are the only collateral, resulting in worse tax treatment.
Instead of paying for the shares with cash now, I agree to pay for them in 10 years, paying interest at the minimum rate the IRS will allow (~2%). The 51% recourse means that the shares themselves are the only collateral for 49% of the loan amount (to limit my risk if the company goes bankrupt and a creditor tries to actually collect on the note).
Another option that I successfully negotiated for is purchasing shares outright at fair market value using a 51% recourse promissory note due in 10 years at the IRS minimum interest rate.
This avoids the exercise window and acquisition concerns, is pretty tax favorable, and largely aligns your treatment with the founders. It is a bit riskier even if the company agrees to offset the loan with bonuses over time, but at lower valuations I think it is worth considering.
In any case, it's worth it to have a good lawyer look over all your options paperwork to make sure you are getting a fair deal.
This exists for high risk merchants -- there are a number of offerings like: https://www.easypaydirect.com/special-needs/load-balancing/
I'd need to do more research to find the best solution but it looks like payment processor risk is definitely something that can be mitigated.
Did the contract in question have copyright to the work only transfer upon payment in full? That is a key piece of leverage that you should include in every contract. This way if the client continues to use the code you can inform them that it is copyright infringement, potentially send the web host a DMCA takedown notice, or have your lawyer pursue them for copyright infringement.
That said, the "self-help" takedown sounded problematic -- much better to have the option of a copyright infringement claim.
If a client pushes back on this provision you can explain that you are 100% insistent but it only comes into question if they don't pay -- that has resolved all objections in my experience, and if they pushed further I'd treat it as a major red flag.
Specializing is the best thing you can do to get higher rate work. Read https://philipmorganconsulting.com/the-positioning-manual-fo... for a lot of great advice on this.
Do fixed fee if you can -- this gives you an incentive to invest in becoming more efficient and gives your clients predictability. Jonathan Stark has some good material on moving away from hourly: https://expensiveproblem.com/
The thing that's been a huge success for my freelancing business and helped me avoid feast or famine is having ongoing, hands-off sales processes that keep going no matter how busy I am with client work.
I hired a VA and originally had her email relevant leads from lead newsletters like LetsMakeApps or Workshop, and now am shifting towards cold email towards relevant leads. Use a CRM (close.io is my personal favorite) and a drip email service like prospect.io for automatic followups -- persistence boosts conversion rates massively for this sort of sales. Block off a day or two a week to take sales meetings and use Calendly to schedule -- meetings popping up whenever is huge impediment to flow, but keeping up sales regardless of how busy you are. If you get overwhelmed, raise your prices to reduce the amount of work you are selling.
Happy to talk about building a freelancing business -- email in profile.