AI Uplift for Careers (Free 80,000 Hours Career advising without an application)

These AIs are getting pretty smart. Maybe smart enough to help you think about your career? I was skeptical when first pitched on the project before joining 80,000 Hours. Turns out AI progress is a good train to hook yourself onto and AI has gotten pretty good at career advice.

It has some advantages compared to a human advisor:

• The AI will spend as much time with you as you need
• It can start now, without a delay
• It can know about all the jobs and opportunities on the Job Board, even the niche ones
• It can work with you over time to build up knowledge of what jobs you’re thinking about and what skills you have
• Telling it more about yourself can, if you choose, add data to our headhunting database. 80,000 Hours has a team that searches this database for hiring managers, including for non-public roles. We’ll say more in a forthcoming post.

(Humans still have their own advantages and you should still apply to advising or talk to your friends.)

Introducing the 80,000 Hours AI advisor

What you’re getting is similar in some ways to a regular chatbot. It’s currently powered by Claude Fable under the hood. It also has:

• Job Board opportunity search
• A career quiz
• Storing info, forming memories, etc.
• Proactive emails
  • If you’re like me, you don’t actually like thinking about your career. I wouldn’t really want my husband nagging me about jobs, but a bot who knows about deadlines telling me about them would help me stick to things.
  • If you’re not like me and you’re excited / don’t find it aversive to work on your career (It seems like there are people like this, based on my user interviews. I was shocked.), then these emails can still be a proactive input.
• As mentioned above, as you build up your profile, you also are (if you choose) telling the 80,000 Hours headhunting database about yourself. It’s much easier for our headhunting team to match up a role to a richly detailed view of someone’s goals and interests than it is to a LinkedIn profile.

Some nice things people have said:

• “I wasn’t accepted for advising from 80,000 Hours when I applied, which was understandable but still disappointing. I appreciate that the AI Advisor makes 80,000 Hours’ approach to career advice accessible to more people!”
• “In approaching my career transition, I found the AdvisorBot to be tremendously helpful for early-stage brainstorming and organizing my thoughts. I only wish I’d known about it earlier in my process!”
• An average rating of 6.0/7 from 911 responses.
  • It shows up after the advisor’s 4th message and asks “Thanks for trying our AI advisor! How are you finding your experience?”

As I say nice things about it, I should also mention that I have seen it write some things that are not what we are aiming for (e.g. poorly calibrated). It’s still experimental and ultimately quality control is hard. Treat it like a smart intelligence, but I recommend talking to a friend or mentor about conclusions it's helped you reach.

The Young Lady’s Illustrated Primer

In Neal Stephenson’s 1995 novel The Diamond Age there exists a “book” which is interactive. It serves as an adaptive pedagogical tool capable of teaching any subject. It is designed to help the main character Nell learn while also leaving her with the intellectual tools to be an independent thinker.

Without having read the book, I’ll stop there. My thesis is that as AI turns to be more and more central to the way we explore the world, I’d like to have a good AI-first way to explore 80k’s content, which is now quite a lot for one person to read / listen to / watch all of.

I’d like that AI to be able to meet you where you’re at, help you with what you want to achieve, and not try to tell you that 80k has all the answers. I’d like to build (and think we can build) the best jumping-off point for impactful careers.

FAQs

Why not use [my favorite way to talk to a generic AI]?

80,000 Hours does not have a monopoly on calling Fable, and “advice from a frontier model” is a large part of the value proposition. claude.ai is a high bar to compete against.

I think our biggest advantage is that we’ve spent a lot of time trying to make sure that it will make an honest attempt to evaluate your fit for a role instead of always telling you you’re a great fit for a role. We have an “eval” (as the term of art in AI product engineering goes) for “the rationally pessimistic user.” We present scenarios of users correctly suggesting that a job might be a bit too much of a stretch for them right now, and make sure the advisor doesn’t provide meaningless encouragement.

Additionally, when I use Claude frequently for a specific domain, I often spend a while setting it up for that use-case. And we’ve spent a lot of time giving our AI advisor the context and guidelines it needs to be helpful in career thinking and job searching.

(And I’ll add yet another pitch for joining the Headhunting database.)

The biggest downside to using our AI advisor instead of your own AI setup is that you may have worked hard to personalize your AI. In the future we might build an integration/MCP setup to get the best of both worlds. Let me know if that is something you’d be excited to use.

Ultimately, try both! Let me know what it’s comparatively good or bad at.

Will it push me towards AI safety?

In my experience, if you ask it to help you find a job in animal welfare, it’ll help you find a job in animal welfare. If you ask it, “What should my cause prio be?” it’ll (discursively, still) raise AI safety.

How’s it made?

I get a lot of people asking if we finetune it. No, Anthropic doesn’t let you finetune Fable, and that’s not generally the best way to get a frontier model to do well at something complicated like free response career advice.

We currently use Claude Fable, but we might switch as better models come out. We don’t do finetuning as we think using existing frontier models provides better performance on complicated tasks (like free response career advice) than finetuned open models.

Perhaps more surprisingly, it doesn’t have something like a RAG pipeline to grab context from 80k’s content. It just …knows the content because it’s part of its training data and because I think the EA way of thinking has just really had a large impact on at least one corner of the internet. It is still instructed to search for content when details or recency matter.

Feedback, etc.

This post was written by JP Addison and does not necessarily reflect the views of 80,000 Hours. I’m only one member of the team of me, Luca De Leo, Sarah Cheng, and Director of Career Services Huon Porteous, who are all working on this product, along with a team of AIs.

So much of what we build, we build in response to user feedback. Please DM, comment, or use the feedback form on the left-hand side of the site.