Personally I think LLM benchmarks make agents worse. All these companies chase the benchmarks, overfit, and think being able to cheat at the math olympiad is gonna get us to AGI. Instead researchers should peer in and get me an agent that can reliably count the number of "i"'s in mississippi.
I don't quite think they cheat at math olympiads, but obviously there are blindspots for the unspectacular tasks. That being said, Mississippi is both a good and a bad question to ask. On the one hand, it's "the bare minimum" to require, on the other hand, is it really a feat? Like, most models can write a piece of code that would compute that. If you show me a task I'm not designed to solve (like count the number of i's in this text), the smart thing is actually to write a program to count them (which LLMs can do).
The best way to measure intelligence is probably to have a model know its strengths and weaknesses, and deal with them in an efficient way. And the most important thing for eval is that ability.
The man lived at the same time my great grandfather did, a ww1 Jewish German veteran.
Everyone’s views on their integration into German / Gentiles society were completely upended throughout those years. Einstein wasn’t always an ardent Zionist, he wasn’t always a pacifist, his lived experiences caused changes. Your question compresses 30 years of life, say 1925-1955, into a simple binary to try and drive some nasty conclusion.
I already responded to you at https://news.ycombinator.com/item?id=39169987 but this is so bad that I need to add something: if you post like this again to HN, we will have to ban you.
Thanks! I wasn't aware of that model; they must be new. Looks like they were introduced back in March of this year. Good to see the barrel hinges. But a couple of questions:
Are the lenses glass? (It says they're G-15, which is supposed to be mineral glass, but I don't know if RB has changed the definition somewhere along the line and replaced them with plastic).
Are these only available in EU, or in the US as well?
G-15 is just the tint color (the gray/green). It’s not the material. I’m pretty sure that only way you’re getting glass lenses from Ray-Ban is if you get the high index prescription lenses.
As a kid I ate this stuff up. In the eighth grade, I defaced my middle school website.
The IT person easily figured out it was me and then tricked me into thinking I would be expelled within days. She pulled me out of class, told me such in the hallway, let me return to class where I held in tears until the end of the day.
Nothing happened and the school year ended a few weeks later. Towards the end of the summer I realized it had been a bluff and I wouldn’t be punished. Took me a few years later to realize how much of a favor that all was! The county school of conduct clearly said cybercrime was punishable by expulsion so she could have absolutely put me in some kind of hell. The fear set me straight hah.
> The IT person easily figured out it was me and then tricked me into thinking I would be expelled within days.
Similar. I wrote a program to emulate a the logon text on a PDP-11 terminal in high-school in the mid-80s and steal a bunch of student passwords. Didn't do anything with them. They were like "trophies."
Nevertheless, the computer teacher found out and had mercy on me. He gave me a project to work on to help him compile stats on a student survey. He was a nice guy.
I did the same thing, only my program pretended to be a DOS-based Novell Netware login screen.
It was just a simple QBASIC program (that's all that was available on the Computer Room machines) running under my own login, which would write usernames and passwords to a text file in my user directory. I figured that I'd harvest a few passwords until someone got frustrated enough to call for the IT admin, at which point he would try to log in and reboot the PC when it failed, apparently "fixing" the problem and erasing any evidence of my dastardly crime.
I was right, and for a few glorious days I got away with it... until one particular arsehole picked on my best friend during recess, and I used his stolen credentials to log into his account and trash his files.
Long story short, I ended up getting expelled, which by a curious confluence of events put me on an unorthodox path that completely changed my life. Funny how things turn out.
> until someone got frustrated enough to call for the IT admin, at which point he would try to log in and reboot the PC when it failed, apparently "fixing" the problem and erasing any evidence of my dastardly crime.
This was precisely my logic as well.
> put me on an unorthodox path that completely changed my life.
I had a similar thing happen. I distributed some malware I wrote on the shared drive and had some people run it (it was extremely basic, just locked people out of the computer with no recovery by taking advantage of how locked down they were; but people lost a lot of work). My programming teacher, who was already dealing with me being a distraction in class, went to bat for me so I didn’t get strongly punished but made me clean it off the drive continuously; other students kept putting it back, so I had to monitor for it.
Not in an urban area. E-bikes fly through sidewalks, bike lanes, crosswalks in NYC. Threatening pets and people at all times. Bikes were already bad, now it’s just more and faster.
Therapy, meds, support groups on Reddit, and exercise are the best thing you can do to deal with childhood trauma. It actually seems to get worse if you don’t manage it, I probably only really started attacking my trauma at 28/29. Made a big difference in personal and professional success.
Yep. I’m personally skeptical of so many other use cases for LLMs but CoPilot is fantastic and basically just autocomplete on rocket fuel. If you can use autocomplete, you can use CoPilot super effectively.
I almost always turn autocomplete off except in circumstances where the API has bad documentation. I also found that copilot was an aggravation more than a help after using it for a couple weeks.
We programmers enjoy writing code. We derive satisfaction when a code is perfect and elegant. But its going to end very soon. Artists are freaking out because things that takes them days to create now only take 2 seconds. We are next.
The writing is on the wall. Programming as we know it is going to end. We should be embracing these tools and should start moving from software developers to software architects role.
Oh absolutely. Twice I made the mistake of coming onto a startup to fix the “mess” the previous dev shop had made.
Both times the founders had developed an incredibly negative relationship with their dev shop. Hell, the company with the Series A was withholding payments, the preseed company at least paid their bills.
I naively thought “of course they hate working with a dev shop, the incentives are wrong. I’m FTE with percentage points equity. I’ll fix it.”
Of course, things got better, but growth stalled or never happened, blame gets shifted, suddenly that founder who was excited to hire top tier talent is very disappointed in the very equity and salary expensive engineer you are. You end up being the new target of their animosity. You start to empathize with the folks in India, China, Ukraine or Argentina whose history lingers in the git blame. They’re not stupid, the incentives were just wrong!
The thing is…the CEO who failed to find a technical co-founder, retain them, or replace them with the same equity agreement has already demonstrated their inability to work with technical people. If they are coming from tech, the fact that they couldn’t convince a former colleague to work with them means they’ll be a terrible manager. If they come from finance, as many founders do, you are a line item.
At the end of this, hopefully sooner than later, you’ll quit, you’ll get fired, or you’ll collect paychecks while looking for the next better thing.
> If they are coming from tech, the fact that they couldn’t convince a former colleague to work with them means they’ll be a terrible manager.
I kind of disagree here. I have been a manager at F*NG and just below those places in terms of pay, and getting my guys to jump to a startup without market fit, for the pay that those types of places tend to offer, I don't see how they would make the jump. The product/idea would have to be VERY enticing for this to happen, while for me personally, a CTO title would be enticing enough to maybe roll the dice on something.
Well, in this scenario, you are actually the 'former' colleague getting recruited to be a co-founder while the other folks stay stable. You just admitted you would and I wouldn't be surprised if there is a PM or even another engineer who would want to join forces. That's different than a complete stranger recruiting you -- I would be very wary.
This was a real scenario I faced and the company/founder was a complete stranger to me- I forget if he just found me on LinkedIn or it was through angel list- I would not have been a cofounder just their CTO. Either way I ended up taking another offer with somewhat lower risk- they were extremely successful but turnover there was on the order of 25% per year. And I did end up taking some people with me and lasted about 5 years there so it all worked out.
I check in to see if the other company ever made it anywhere and nothing ever comes up so I guess I wasn't missing out on startup riches.
Be nice but remember you have objective skills and you need to suss out the CEOs objective skills.
1) Can they raise money? Proof is good here. Sometimes someone will say "I want you as my co-founder to go and raise my seed round" and you can condition leaving your current job on getting at least some checks in. Remember, this founder _knows_ you can deliver the goods, you are an employed software engineer. Can they?
2) Can they recruit? The process might be a mess but are they treating your time and concerns with respect?
3) Do they view you as an equal partner? Make sure you nail down exactly what they want. If they want a true co-founder, ask for 50%. If they don't, make sure you understand what it means to be an employee to a non-technical founding team.
