Igor Korostil (@eeghor) header
Igor Korostil (@eeghor) profile picture

Co-Founder & CTO @ Mirmi AI | AI Strategy, GenAI & Automation

Sydney, New South Wales
Joined July 2016
Photo by Igor Korostil (@eeghor)Photo by Igor Korostil (@eeghor)Photo by Igor Korostil (@eeghor)Photo by Igor Korostil (@eeghor)Photo by Igor Korostil (@eeghor)Photo by Igor Korostil (@eeghor)Photo by Igor Korostil (@eeghor)Photo by Igor Korostil (@eeghor)Photo by Igor Korostil (@eeghor)Photo by Igor Korostil (@eeghor)
Igor Korostil retweeted
The Illusion of Thinking in LLMs Apple researchers discuss the strengths and limitations of reasoning models. Apparently, reasoning models "collapse" beyond certain task complexities. Lots of important insights on this one. (bookmark it!) Here are my notes:
136
601
128
4,382
956,778
Igor Korostil retweeted
Yet more evidence they people can’t accurately detect well-prompted AI writing and AI can’t accurately detect well-prompted AI writing, either. arxiv.org/pdf/2407.08853
30
182
22
883
87,372
Here, in full directly on Twitter, is "A Hackers' Guide to Language Models". This 90 minute tutorial is designed to be the one place I point coders at when they ask "hey, tell me everything I need to know about LLMs!" It covers both @OpenAI models and open source ones in depth.
77
998
52
5,912
777,515
Igor Korostil retweeted
ChatGPT plugins are being rolled out to all Plus users this week. This could be a game-changer as big as the App Store. Check out my top 10 favorite examples so far:
68
242
18
1,565
814,157
Igor Korostil retweeted
New paper: On the unreasonable effectiveness of LLMs for causal inference. GPT4 achieves new SoTA on a wide range of causal tasks: graph discovery (97%, 13 pts gain), counterfactual reasoning (92%, 20 pts gain) & actual causality. How is this possible?🧵 arxiv.org/abs/2305.00050
32
274
54
1,315
457,146
Very excited to share some early discussion of how we're using Bayesian Optimization to tune Lyft's marketplace using time-randomized experiments. Much more on the methodology to come! eng.lyft.com/parameter-explo…
5
42
3
353
Igor Korostil retweeted
#Hollywood’s declining creativity? Proportion of original #movies in the worldwide top 50 grossing films (1978-2019). #dataviz source: reddit.com/r/dataisbeautiful…
3
10
2
43
Igor Korostil retweeted
Baker-Miller Pink is a color that reduces hostile, violent or aggressive behavior in some settings. https://teapawt.net/t.co/0OPxgAZeo0
1
2
1
8
Igor Korostil retweeted
An Introduction to Biomedical Image Analysis with TensorFlow and DLTK https://teapawt.net/t.co/09EygpcGW4 #AI #DeepLearning #MachineLearning #DataScience
14
29
An interesting research article on the prediction of individuals’ Sixteen Personality Factors (16PF) by analyzing their facial features.

3
38
12
63
Igor Korostil retweeted
Hallucinogenic Deep Reinforcement Learning using Python and Keras https://teapawt.net/t.co/wjcl5YYqiG #AI #MachineLearning #DeepLearning #DataScience
3
1
14
Igor Korostil retweeted
Command Line Tricks For Data Scientists medium.com/@kadek/command-li…
69
5
159
Igor Korostil retweeted
Representing Language with Recurrent and Convolutional Layers hergott.github.io/language-r…
1
1
Igor Korostil retweeted
Generating Drake Rap Lyrics using Language Models and LSTMs buff.ly/2qi9hYk
DeepPavlov: open-source library for building chatbots, built on top of TensorFlow & Keras github.com/deepmipt/DeepPavl…
4
269
10
736
Comparing #Python Clustering Algorithms (HDBSCAN docs): hdbscan.readthedocs.io/en/la…
7
49
Igor Korostil retweeted
“Using Word2vec for Music Recommendations” by @ramzikaram towardsdatascience.com/using…
2
1
Igor Korostil retweeted
Customer #Analytics: Using #DeepLearning With Keras To Predict Customer #Churn https://teapawt.net/t.co/O2jCb75NE3 #CX #CustomerSuccess #datascience
5
3
Igor Korostil retweeted
Goodbooks-10k: a new dataset for book recommendations fastml.com/goodbooks-10k-a-n…
17
27