ammon

Tabula Bio | Founding Machine Learning Scientist | San Francisco, CA | ONSITE | Full-time

We use generative ML to design bacteriophages that treat antibiotic-resistant superbugs (bacterial infections). Using phages in this way is not a new idea. Phages already cure patients in research contexts today. But the process is too slow for the ~50K Americans who die each year from antibiotic-resistant infections. Our models make phage therapy fast, and we’re working with real patients today. Join our ML team. Biology experience is not required; what matters is strong machine learning skills and interest in genome design and saving lives with computer science!

Email me at ammon@tabulabio.com

And we know what the adaptation is: calorie constraint. We evolved in a calorie constrained environment. We don't live in one now. Our set point for desire to eat is clearly too high. None of this means that glp-1 inhibitors don't have other side effects, of course.

Nope, LLMs are quite functional in non-english languages. My partner regularly works with ChatGPT in Turkish

For one thing, viruses keep bacteria in check. Bacteria populations live in an equilibrium standoff with bacteriophages (viruses). The human gut contains more bacteriophage virions than bacterial cells. So eliminating all viruses could lead to bacteria overgrowth

Indeed. It is common to not have changed your mind about the thing someone is upset about, but not feel good that they are upset. There needs to be a way to communicate that

Most employers on Triplebyte (90%) are hiring in the US currently. However, we're working on something that us going to change that (announcement coming soon).

As of 2 years ago, every profile on Triplebyte is not visible to recruiters unless an engineer explicitly makes it visible.

That was true of our process as it existed in 2017/2018 (and was a part of why that business was not viable). At this point, what we do is develop tests (backed by psychometric models). These are more accurate than human phone screens (and especially more accurate at finding people who have strong skills bad "bad" resumes)

I’m the CEO of Triplebyte. I just looked into this, and we are not automatically making anyone's profile public without their consent. OP, it looks like you had set your profile to be visible some time ago (I can provide details via email but won’t here for privacy purposes), and just hadn’t gotten a message until now. The support response you received was incorrect; you weren’t affected by any recent changes to the site.

Since there wasn’t anything broken on our end, there shouldn’t be anyone else impacted by this. But as part of making sure that OP wasn’t visible by mistake, I had my team double-checked to make sure our previous fix from last year was properly retroactive. It was.

More generally, I don't think that tricking anyone is a viable way for us to run a business. We’re trying to create a marketplace that can open opportunities for engineers who wouldn’t otherwise have them, and we need the trust of engineers in order to do that.

But how much more important? :) Sorry, could not help myself.

That's exactly what the article we're all commenting on is about!

We experimented with this, as well as with trial periods (which are sort of the same thing, but with a time limit). Trial periods ended up being easier to think about and optimize (each account either converts or not at the end of the period). But we may revisit this in the future.

We do have discounts for smaller startups (under $10m in funding). Happy to talk about that if you're interested. ammon@triplebyte.com

So, under the old model, order 200k engineers applied to us. Because we were exclusive gatekeepers, around 3% of those were "accepted" onto the platform. Around 2/3 of accepted candidates received an offer, and around 1/2 of offers were accepted.

The main problem we had with the contingency model is that it resulted in companies using us for a few of their "hardest" hires, rather than as a bread-and-butter part of their recruiting. For example, almost no one wants to pay a contingency fee on junior or remote hires (because they are perceived as 'easier'). This made our platform kind of suck for junior and remote engineers. The subscription model is more of a commitment from a company (I understand why some companies don't choose to use us.) But it means that once a company signs up, they want to make as many hires as possible through us.

I generally agree with the phrase "if you're not paying for a product, you're the product". But the market for engineers is just so lopsided that I think it's less true here. There are a lot of recruiting companies. The only real thing that sets one apart from others is whether they have candidates. So one way to look at it is that yes, we are incentivized to build what companies want, but the main thing they want is for us to have engineers. And the only way we get engineers is by building what engineers want.

That does not fully express my motivations. I am an engineer and find the idea of making the process better for engineers more exciting than making it better for companies. But it explains how I think the incentives work.

I'd say that main value was that we opened doors for people (got them opportunities they would not have been considered for without us). I don't think what we need to be gatekeepers to do this. Yes, not everyone can succeed (get a job at at top company). But we can help everyone show their skills in the way that's best for them. We can fight ghosting and lying and create a less hostile process. We're keeping our quiz (so that people who do well on tests can get opportunities that way), and also creating a job search process for people who do not want to do a quiz (the majority of engineers). The idea is that they will show their skills other ways (past experience, side projects, open source work). By us not being a gatekeeper we open these other paths.

Fair comment. I certainly don't think of us as a market leader. LinkedIn is the market leader. Honestly, the last year has been pretty hard for us (COVID and the problems with our model that I talk about in the post).

That said, I think a hiring process that puts engineers in control (no ghosting, get data on when a recruiter looks at your application, search ranking by whether companies lie to candidates) is something that should exist. I want to try to build it!

Companies do pay us. But I think that our real long-term incentives still pull toward building what engineers want, not what companies want. I think that platforms like LinkedIn and Indeed have just gotten this wrong, because they focus on all jobs (not just engineers), and because demand for engineers is stronger than it has ever been. Take a look at this thread from last week:

https://news.ycombinator.com/item?id=27501675

Our bet is that LinkedIn is going to fragment. They are just not creating a hiring process that most engineers like. People tend to either get ghosted, or overwhelmed with low-relevance inbound (almost no one gets the "right" amount of attention). Companies need to go where the best engineers are, not the other way around. So I think our long-term incentive is to fix these problems.

In any case, I'm committed to giving this a try. There is danger that we get pulled toward building for companies. I want top guard against this by being public about what we're doing, and "showing our work" as we go.

This is a perfect example of why we're making these changes (and the problem that came from us being a gatekeeper). There are lots of different ways to show skill. We don't want to be in the position of deciding who "deserves" a job.

There's some risk of that. But I think that we can make it work. Basically, the model I have is that there are many different ways that an engineer can signal competency to a company and get a job. One way is doing well on our quizzes. But another way is having impressive experience at a good company, or a lot of open source contributions. Any one engineer might be really good, but look bad on one of these metrics (someone who gets stressed and does poorly on tests, or someone who has mostly worked for the government and does not have side projects). This happens a lot. When we ONLY ranked engineers by our assessments, we just missed people. The goal now is to let people show skill in any of these ways.

So, we ARE keeping the quiz (and companies still trust it as much as they did in the past). It's just not mandatory.

I'm really glad that the old process worked so well for you. When it worked, it was pretty great. We are losing something with this pivot. The problem is that the old process did not work for far more people than it did. For every wonderful experience with a TM there were 10+ engineers who we screened out. That’s why we made this change.

The difference is that I think we're in a position to actually change behaviour (require companies to not do bad things like ghost candidates, lie about salary).

I hadn’t thought about this from a legal contract perspective. I like the idea of making our commitment enforceable. Do you have ideas of what you'd want in a contract?

This is certainly an issue. As you say, it’s a problem on yelp and glassdoor. It’s why most nightclubs have low yelp ratings (people they exclude give them low scores). There are a few things I think we think we can do:

1) Look at relative data, not raw ratings. Written reviews or raw numerical scores are what show the nightclub effect the most. What we care about is a ranking of companies (showing the better companies first in search results). This may not be as impacted by the problem. If sour grapes from people who failed interviews overwhelm other signals, we can only look at the scores from people who pass (or normalize the two groups separately).

2) Ask very specific questions (and maybe provide dummy options to attract sour grapes). I suspect that questions like "did this company say they could meet at $250 salary and then offer you less?" will get more accurate answers than "does this company have a toxic culture?". (We do actually want to get culture data, but I want to be really specific about the culture questions we are asking).

3) Use objective facts, not subject opinions. For example, we can tell when a company ghosts a candidate (because they reply through a proxy email that we control). So we're listing the 'ghost rate' for each company.

Ok, we’ve looked into this. The problem was a bug in our reminder email. We only intended to send the reminder 3 times, but it was going out every 6 weeks forever. We’ve turned off all reminder emails for now (and will audit them after the holidays). Had I realized this was happening sooner, we would have fixed it sooner. I’m absolutely NOT trying to spam / growth hack our way to anything. I don’t think that would even work. Our goal is to help people who are strong programmers but don’t have credentialed resumes get jobs. We only succeed by helping engineers. It’s clear from some of the replies here that I’ve lost the trust of a part of this community. I feel pretty bad about this. I’m going to do an Ask HN post next year and get everyone input on what Triplebyte should do to help engineers and not be spammy. If anyone has any thoughts now and would like to talk, please email me at ammon@triplebyte.com. I promise to listen to anyone (I’d especially like to talk to people critical of us). Have a wonderful Christmas, everyone!

We don’t have shadow profiles or anything like that

Fixed

Well, it's painful for me to read that. Is there any chance I could get you on the phone to talk about what we're doing and how to make it not scummy? (That offer goes out to anyone reading this). Email me at ammon@triplebyte.com, and I'll find a time to talk. Our mission (helping people get jobs based on the job skills they have, not what their resume says) is something that I think is really important.