5 ms·

I've had massive success with this in ML. My twitter DMs have become almost-daily office hour sessions with people trying to solve hard ML problems. The last fe

I've had massive success with this in ML. My twitter DMs have become almost-daily office hour sessions with people trying to solve hard ML problems. The last fellow was so happy he venmo'd me $250 after I merely asked for "pay whatever you thought that was worth." Apparently putting "DMs open" in your twitter bio is an excellent way to meet new people. Also, tweet (often!) about what you're working on. Post lots of questions, notes to self, and screenshots or videos. To get followers, you'll need to go inject yourself into conversations. That seems to be the most reliable way to get noticed. We also run an ML discord server and do informal office hours for people working on various ML products. It may sound cheesy, but I try to emulate what pg would do in a YC office hours session. It seems to keep them pretty focused. So, yes! Glad to see office hours are becoming mainstream. Try it out; you'll probably be surprised.

arch-ninja 6y ago
It's kind of neat observing the same economics that drive Instagram influencers also drive sharing logical thinking. I wonder if both sides of the human brain behave the same way? (organic v. ogic). Actually it's far more likely that social effects are the cause, and social effects benefit both organic runaway trains as well as logical runaway trains.
- an_opabinia 6y ago
  On the other hand, $250 was probably way too little money. I question the wisdom of helping people who are professional be-at-the-right-place-at-the-right-timers. That includes people working for giant companies messaging open source maintainers, or any number of total inequities.
  - ianlevesque 6y ago
    What if you want a referral into one of those giant companies someday? Can't hurt to know a guy...
    - kyawzazaw 6y ago
      you could just write a post on teamblind.com if you are that level or just make a tweet
      - dylan604 6y ago
        you're putting a lot of weight on someone having the right followers or someone casually reading some rando.com website. comparing that roll of the dice big bang level of random vs having direct contact with someone that uses code you created and can vouch for you seems skewed in the wrong direction to me.
        kyawzazaw 6y ago
        yeah but that's how many have gotten jobs. It's just a matter of ROI. Wrong direction for whom?
victor9000 6y ago
What are some of the more interesting problems you've encountered using this approach?
- sillysaurusx 6y ago
  I helped a fella figure out how to imbue their GPT model with emotional awareness. That was one of the cooler ones, and we solved it in about an hour or so. The actual technique will take a month or so to execute, but the core idea is pretty much guaranteed to work.
  - victor9000 6y ago
    Interesting! Are you able to share what you came up with? It seems like at minimum you could verify that the sentiment analysis for your generated text matches that of the input prompt. I've been working on getting GPT2 to write poetry and I've made a fair amount of progress by fine-tuning on a poetic corpus. Then during generation I constrain tokens that would break the style of the previously generated text. It has already produced a few samples that are surprisingly moving. Here's one of my favorite cherry-picked samples which was generated with a prompt of "The sea grew": "The sea grew angry, and turned to fire, And all the stars did mourn for earth and sky. The wind did moan, as when one who hath lost Love through some evil counsellor, hears A lamentable voice that sobs and sighs; Or when some old sorrow's semblance doth appear" It makes me think of someone describing the meteor strike that killed off the dinosaurs. To achieve this, I simply perform analysis on the tokens as they are generated, then turn off the logits for tokens that would break the current flow. I'm currently working on enforcing meter and rhyme using a similar approach. I've also had success with creating checkpoints by pickling the model state at a good stopping point, like at the end of a line, then letting the model do its thing and reverting back to the checkpoint if it produces something that doesn't match what I'm looking for.
    - sillysaurusx 6y ago
      I might be able to, but I'll have to ask. Want to DM me on twitter, so that I can share it with you (if permission is given) without broadcasting it to the world? https://twitter.com/theshawwn https://twitter.com/theshawwn
iainctduncan 6y ago
I'm stealing that too. "DM's open" coming right up.
carols10cents 6y ago
"DMs open" in your twitter bio as a woman is an excellent way to get lots of "hello" messages from randos...

arch-ninja 6y ago

It's kind of neat observing the same economics that drive Instagram influencers also drive sharing logical thinking. I wonder if both sides of the human brain behave the same way? (organic v. ogic). Actually it's far more likely that social effects are the cause, and social effects benefit both organic runaway trains as well as logical runaway trains.

an_opabinia 6y ago
On the other hand, $250 was probably way too little money. I question the wisdom of helping people who are professional be-at-the-right-place-at-the-right-timers. That includes people working for giant companies messaging open source maintainers, or any number of total inequities.
- ianlevesque 6y ago
  What if you want a referral into one of those giant companies someday? Can't hurt to know a guy...
  - kyawzazaw 6y ago
    you could just write a post on teamblind.com if you are that level or just make a tweet
    - dylan604 6y ago
      you're putting a lot of weight on someone having the right followers or someone casually reading some rando.com website. comparing that roll of the dice big bang level of random vs having direct contact with someone that uses code you created and can vouch for you seems skewed in the wrong direction to me.
      - kyawzazaw 6y ago
        yeah but that's how many have gotten jobs. It's just a matter of ROI. Wrong direction for whom?

victor9000 6y ago

What are some of the more interesting problems you've encountered using this approach?

sillysaurusx 6y ago
I helped a fella figure out how to imbue their GPT model with emotional awareness. That was one of the cooler ones, and we solved it in about an hour or so. The actual technique will take a month or so to execute, but the core idea is pretty much guaranteed to work.
- victor9000 6y ago
  Interesting! Are you able to share what you came up with? It seems like at minimum you could verify that the sentiment analysis for your generated text matches that of the input prompt. I've been working on getting GPT2 to write poetry and I've made a fair amount of progress by fine-tuning on a poetic corpus. Then during generation I constrain tokens that would break the style of the previously generated text. It has already produced a few samples that are surprisingly moving. Here's one of my favorite cherry-picked samples which was generated with a prompt of "The sea grew": "The sea grew angry, and turned to fire, And all the stars did mourn for earth and sky. The wind did moan, as when one who hath lost Love through some evil counsellor, hears A lamentable voice that sobs and sighs; Or when some old sorrow's semblance doth appear" It makes me think of someone describing the meteor strike that killed off the dinosaurs. To achieve this, I simply perform analysis on the tokens as they are generated, then turn off the logits for tokens that would break the current flow. I'm currently working on enforcing meter and rhyme using a similar approach. I've also had success with creating checkpoints by pickling the model state at a good stopping point, like at the end of a line, then letting the model do its thing and reverting back to the checkpoint if it produces something that doesn't match what I'm looking for.
  - sillysaurusx 6y ago
    I might be able to, but I'll have to ask. Want to DM me on twitter, so that I can share it with you (if permission is given) without broadcasting it to the world? https://twitter.com/theshawwn https://twitter.com/theshawwn

iainctduncan 6y ago

I'm stealing that too. "DM's open" coming right up.

carols10cents 6y ago

"DMs open" in your twitter bio as a woman is an excellent way to get lots of "hello" messages from randos...