Alex Sherstinsky

Scientist, Engineer, Manager, Musician, Athlete, Thespian, Father, Husband, Son, Friend

[Alex Sherstinsky @AlexSherstinsky]

Fundamentals of Recurrent Neural Network (RNN) and Long Short-Term Memory(LSTM) network -- https://t.co/jZWEtxQCjQ (published on January 29, 2020 in Elsevier "Physica D: Nonlinear Phenomena"; also available at https://t.co/QiUl6X72I3).

[Alex Sherstinsky @AlexSherstinsky]

The science of #ConversationAnalysis allows analyzing talk patterns without having to listen to the entire recording or reading the entire transcript. Based on this analysis, improvements can be made to make conversations more effective. https://t.co/LIE8NBkJXs

[Alex Sherstinsky @AlexSherstinsky]

@maxmbeck Humangous congratulations! So excited to learn from your thesis report, papers, talks, and other materials!

[Alex Sherstinsky @AlexSherstinsky]

@askalphaxiv I have been wondering about trying this for years! The inspiration for trying this has been the Hopfield Network, which excludes self-connections. It always seemed like bias to me (of course the token will be most strongly correlated with itself!). So glad to see this result!

[Alex Sherstinsky @AlexSherstinsky]

@zhaisf Thank you for sharing; is the full content available?

[Alex Sherstinsky @AlexSherstinsky]

Claude Code has been 'ruminating,' 'deliberating,' 'pondering,' 'reflecting,' and 'strategizing.'

Never once 'hallucinating.'

Not even when it absolutely was.

#ClaudeCode #AI

[Alex Sherstinsky @AlexSherstinsky]

@CalcCon @Yuchenj_UW You are so right; hoping to make it to the gym today! :)

[Alex Sherstinsky @AlexSherstinsky]

@CalcCon @Yuchenj_UW Charles, I am sorry! I have been consistently guilty! I know you are 100% right, there is just so much to do! :)

[Alex Sherstinsky @AlexSherstinsky]

https://t.co/XMmr7DGe3T

[AGI House @agihouse_org]

Thanks to @GoogleDeepMind and @GraphonAI - we had a lot of fun hosting Gemini 3 build day @ AGI House last Saturday

Cc. @DynamicWebPaige @vadiamit @bonniesjli @JessicaChenX

[Alex Sherstinsky @AlexSherstinsky]

The encouraging part is that regular up-skilling, which must now be part of a standard development cycle, is also more readily accessible -- as self-guided, AI-assisted, enjoyable part of an engineer's work.

[Alex Sherstinsky @AlexSherstinsky]

My reflections on using AI-assisted software development tools thus far is that without an experienced operator, any such tool quickly gets out of control and generates mountains of error-prone, unmaintainable "code" (in quotes, because there is no quality in that "code").

[Alex Sherstinsky @AlexSherstinsky]

This happens much more frequently than before, because the pace of development is so much greater.

[Alex Sherstinsky @AlexSherstinsky]

@YungSungChuang @yoonrkim @jacobandreas He was (maybe still is?) one of the fastest and smoothest players on the ice! I totally remember the group he was in (Speech, in building NE43). :-)

[Alex Sherstinsky @AlexSherstinsky]

https://t.co/XMmr7DGe3T
#ConversationalAI #ConversationDesign #ConversationAnalysis

[Alex Sherstinsky @AlexSherstinsky]

Conversation Analysis reveals what truly unfolds in business conversations — far beyond the binary "won/lost" label common in sales. A powerful reminder that the "best" outcomes often transcend immediate closes, turning apparent losses into stronger, trust-based wins later.

[Alex Sherstinsky @AlexSherstinsky]

@TheVixhal How does this spectral approach incorporate the nonlinear element, which is so important for enabling the correlation based systems including the Transformer to learn complex data patterns? Thank you.

[Alex Sherstinsky @AlexSherstinsky]

@karpathy My amazing former coworker Justin Zhao published a thoroughly researched paper on this very subject: https://t.co/Uo9VAElhgC -- enjoy!

[Justin Zhao @justinxzhao]

The Language Model Council paper came out last year (NAACL), but I want to repromote it in light of:

1. Ongoing interest in LLM evals.
2. More strong models (DeepSeek, Kimi, Minimax) coming from countries outside the US.
3. Growing evidence of political values being explicitly diffused into the latest models (e.g. Grok 4), and increased interest in political bias in AI models (https://t.co/9TEqyUjHgq, https://t.co/2PX6kiiSst).
4. @karpathy's llm-council demo (https://t.co/dgDl1vcIa5)

Paper: https://t.co/op9PxjVSah
Github: https://t.co/PL72K5mTt3
Slides: https://t.co/CAraFTyhkh
Recording on YouTube: https://t.co/xzp7i9gYuf

(1/8)