Berkan Ottlik (@berkott3) header
Berkan Ottlik (@berkott3) profile picture

Machine learning research✌️😎🏝

Joined January 2021
Photo by Berkan Ottlik (@berkott3)Photo by Berkan Ottlik (@berkott3)
I think people think that non-attachment leads to not caring, when in fact it can just lead to continuous motion, which is what actually caring looks like. You can move from moment to moment without sinking into the trap of negative minima or rumination or artless circling.
1
1
1
397
Berkan Ottlik retweeted
Excited to share that my book, The Worlds I See, will be published on Nov 7, 2023 from @Flatironbooks & @melindagates' Moment of Lift Books. I believe AI can help people & I hope you’ll come along on the journey. Preorder momentoflift.com/the-worlds-… #MoemtnofLiftBooks #WorldsISee 1/
84
462
90
2,423
1,092,296
If your loss ever looks like this, it might be because you aren't calling optimizer.zero_grad() in PyTorch!!! Learned this the hard way lol.
1
2
192
Berkan Ottlik retweeted
Big news: we've agreed to acquire @MosaicML, a leading generative AI platform. I couldn’t be more excited to join forces once the deal closes. databricks.com/mosaic-news
31
196
70
1,237
482,725
Unexpected description of GPT4 architecture from geohotz in a recent interview he gave. At least it’s plausible.
14
89
50
527
646,101
Berkan Ottlik retweeted
OpenAI's Discord is literally a prompt goldmine. It's where all the prompt nerds hang out and design the craziest prompts. Here are the 10 best mega prompts I found in there:
104
401
28
3,192
1,263,004
Is the human race just running gradient descent on a mass scale?
1
102
Berkan Ottlik retweeted
Adam, a 9-yr old optimizer, is the go-to for training LLMs (eg, GPT-3, OPT, LLAMA). Introducing Sophia, a new optimizer that is 2x faster than Adam on LLMs. Just a few more lines of code could cut your costs from $2M to $1M (if scaling laws hold). arxiv.org/abs/2305.14342 🧵⬇️
87
596
116
3,460
1,049,576
Anyone know alternatives to @modal_labs that you can run on your own hardware/VPC?
64
Berkan Ottlik retweeted
1/ @openai & @AnthropicAI bills got you down? a new paper from Lingjiao Chen, @matei_zaharia, & @james_y_zou last week shows you can use "LLM cascades" to cut down on cost (and even improve accuracy!!)
14
38
2
276
61,638
Sunday read. This is hilarious. Engineers are competing who creates the worst UI on Reddit. I think the winners are: 1/ Enter your phone number
459
12,611
1,770
62,496
14,663,215
Why does the login to ChatGPT suck and take so long? Bard and HuggingChat are so much better in that respect, which is why I sometimes default to them.
1
1
136
Berkan Ottlik retweeted
Replying to @WSJ
Concerning
697
455
242
10,736
788,048
Berkan Ottlik retweeted
bbe (big builder energy) in nyc today 😃 congrats to all of the super talented hackers & thanks @contrary @betaworks @lux_capital for hosting!
4
2
49
25,867
Berkan Ottlik retweeted
The bravest souls at the NYC gen AI hackathon - 4am and still going strong
2
7
2
34
9,213
@ylecun closing his presentation with some conjectures #phildeeplearning
15
85
20
504
267,496
Yet another blog about transformers: berkan.xyz/posts/2023/01/tra… (except it's written by me and I think it's decent 😊)
3
60
Berkan Ottlik retweeted
Twitter is an anti-compassion machine that stokes our tendency to dehumanize and even hate people we don't know personally. Four little exercises I use to combat that and crank up the compassion in my mind:
204
1,298
243
6,193
Imagine one day we have parents for AI's, similar to the way we have parents for human babies today.
1
Humanities are awesome, I want to emphasize it more to my kids than I have in my life up till now.
2