Can confirm. I did a clean-room implementation of this in Rust just by reading the paper. Will update once they release the reference implementation. https://github.com/ptaranat/zeromem
I use GLM 5.2 via Neuralwatt and it's gotten so cheap I wouldn't mind cancelling my personal Claude subscription if work gave me one. I've spent 374M tokens this month and it only cost me $18 on energy-based pricing.
I went down this rabbit hole last year after buying a carbonator. Rather than mixing a bunch of oils together, I bought my flavors from Bakto Flavors (based in NJ, USA) which is founded by Dr. Daphna Havkin Frenkel who did her research in food sciences and biotechnology, focusing on vanilla. The cola flavor is really good, and I add acetic acid (Vitamin C) + electrolytes to it. If I'm feeling it, I'll add in vanilla, cherry, or lime flavors to it.
Are there any reasonably priced water carbonation systems that aren't sodastream adjacent?
I know there are hacks for the older sodastreams where you can buy a large tank from a foodgas company, but what I really want is a machine in my house that I can bring a cup to and fill with soda without it costing me ~$1,000.
The only cheap way to carbonate water is in a batch process, like a SoadaStream or equivalent DIY method of releasing CO2 out of a straw into water in a pressurized vessel.
If you really want a faucet that dispenses carbonated water, the cheapest method is to buy just the part of a commercial machine that does the carbonating. It consists of a high-pressure tank with a carbonated water output, and separate water and CO2 tank inputs, as well as a pump that pressurizes the input water, to match the CO2 pressure, so that the water can enter the tank.
If you search Craigslist or eBay for "carbonator with pump" you can usually find something for ~$200. Throw in another ~$100 for a used tank, regulator, and faucet, and you can get a 1920's style seltzer fountain in your own home, then add syrups if you want flavored sodas, or throw in some milk and chocolate syrup and call it an egg cream, despite it containing neither.
I once had a conversation with a potential co-founder who literally told me he was pasting my responses into AI to try to catch up.
Then a few months later, another nontechnical CEO did the same thing, after moving our conversation from SMS into email where it was very clear he was using AI.
Small business owners work 997 and you don't see them incessantly posting about it. That's the catch, though. They own the business. Founders can subject themselves to 996 all they want but it's a failure of management to expect that from employees for less than 1% equity.
I took a break from tech to open my own bookstore and I definitely work more hours than when I worked at a pre-IPO $7B startup. I'm way less stressed. At least my bookstore doesn't wake me up at 3am 3 nights in a row, and expect me to come to work the next day.
I interviewed with an early stage pre-seed startup with a very young team, like 25-27. I was interviewed by someone way more junior than me. According to the recruiter, in 3 months, I've made it the furthest and he told me this startup was churning through top tier candidates left and right.
After my interview, I immediately knew why. The team was so junior they didn't know how to evaluate senior talent. They didn't know what they wanted. I've arguably interviewed more candidates than the person interviewing me.
Last I checked, they still haven't filled that role.
The strong hires I've given all came from underrated candidates who didn't come from trendy backgrounds. Still think Dan Luu's advice holds up even more at early stage startups. https://danluu.com/programmer-moneyball/
This is so interesting to hear. From what I've seen, probably half of recent yc startups have founders below 30. I wonder how senior talent views being interviewed by people who are essentially junior/mid developers.
I'm in my 20s with good credentials and have quite a few friends in the startup world. I would never feel comfortable interviewing someone with 10+ years of industry experience.
> I would never feel comfortable interviewing someone with 10+ years of industry experience.
I would say that's probably overcompensating. I've got about 20 years of startup experience at this point, and one of the things that frustrates me the most is a kind of zero-sum mindset, where you "pass" an interview or not.
In the best cases, interviewing is a conversation, a path to better understanding for both parties. The idea that you're "not qualified" is just as silly, in my opinion, as the idea that an hour-long interview lets someone pass judgement. We can both gain, and maybe I'm exactly what you're looking for, in terms of someone who brings skills or perspective you don't have. Maybe it's obvious that I'd be an awful fit. But either way, I believe everyone has something valuable to bring to the conversation.
Some of the best times I've been involved in interviewing, we've had even an intern talk to someone. If they're helpful, clear, and kind, that can be a huge signal. It's kind of a cliche, back in the day, that you ask the office manager how the candidate treated them, but it's absolutely true that if you treat people "below you" in the hierarchy poorly, that's a red flag, to me.
As a senior (in age and experience), it is sometimes not pleasant.
The junior interviewer might be really smart and extremely motivated, but ready to argue about something very specific while missing the forest for the tree.
Years ago, I was interviewed by two young guys at meta. They asked me to solve on a white board a problem to which the obvious and expected solution was a binary search. Which I did.
I wrote a generic binary search function, and then used it in another function. I stepped through the code of each functions line by line as attempt to prove correctness.
They wouldn't have it. They argued I could only prove it was working by stepping through both functions together. While I argued the literal point of using (pure) functions was to simplify by composing and abstraction.
Things got quite heated up. Especially with one of dude. I just left right there and then.
I have been exclusively doing this in the past year, selling my services as “hardening vibe-coded prototypes for production” or “helping early stage startups scale”.
In the best cases, they were able to reach funding or paying users. Architecture debt is one of the worst kinds of tech debt, so if you set it up right, it’s really hard to mess up.
In the worst case, after my contract ended, the CEO fired the whole US engineering team and replaced them with offshore resources. This was an example of messing up despite the architectural and procedural safeguards we built.
To be a nit picky well acktually…one of your articles opens with, “ In the beginning was the Word, and the Word was import this, spoken by Guido van Rossum…”
I believe ‘import this’ was actually penned by Tim Peters.
> This is working well NOW, it will probably not work in 2 weeks, or it will work twice as well. ¯\_(ツ)_/¯
This all feels like spinning the roulette wheel. I sometimes wonder if AI proponents are just gamblers who had the unfortunate luck of winning the first few prompts.
A comparison I've seen isn't to roulette but to a slot machine. Anthropic itself encourages its employees to treat its use for refactors as a slot machine. [1]
It seems like an idea worth exploring formally but I haven't see that done anywhere. Is this a case of "perception of winning" while one is actually losing? Or it it that the winning is in aggregate and people who like LLM-based coding are just more tolerant of the volatility to get there?
The only study I've seen testing the actual observable impact on velocity showed a modest decrease in output for experienced engineers who were using LLMs for coding.
That really resonates. I've found myself questioning whether I'm wasting my time writing a piece of code: what if the LLM could do this more quickly? So I try it, almost every time, and sometimes it does, sometimes it doesn't. Am I really saving myself any work in the long run? Honestly I don't know. I feel like it's just causing me to work more because it feels like a game and that is, ultimately, where the results are coming from.
