Pinned Tweet
Fundamentals of Recurrent Neural Network (RNN) and Long Short-Term Memory(LSTM) network -- sciencedirect.com/science/ar… (published on January 29, 2020 in Elsevier "Physica D: Nonlinear Phenomena"; also available at arxiv.org/abs/1808.03314).
1/2 Nobody calls a pest control company asking to be enrolled in a recurring plan. They call about the ants in the kitchen.
Yet the plan is what the business runs on. And in about 1 in 4 first calls, both options are genuinely in play: a one-time treatment or a plan.
2/2 We analyzed three months’ worth of sales calls in order to understand how the best agents navigate between what the customer is calling to buy and what the business wants to sell. Full details: convoscience.com/blog/nobody…
"I tried it myself" isn't a lost sale. Failed-DIY callers convert best — ~70% above baseline. Agents who ask about the attempt, normalize it, and give honest credit to the effort book above baseline too. Engaging DIY earns the sale:
#ConversationAnalysis
convoscience.com/blog/when-c…
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. linkedin.com/posts/alexshers…
Claude Code has been 'ruminating,' 'deliberating,' 'pondering,' 'reflecting,' and 'strategizing.'
Never once 'hallucinating.'
Not even when it absolutely was.
#ClaudeCode #AI
Alex Sherstinsky retweeted
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
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").
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.
Really excited to attend PyTorch Conference 2025 at Moscone West in San Francisco next week! Looking forward to reconnecting with the ML Engineering community, diving into keynotes, hands-on talks, and real-world use cases. If you're going, let’s connect! #PyTorchConf #PyTorchCon
"You didn't like that movie?" is not really a question in search of a comprehensive analysis of the particular film; rather, the real speech act of that sentence is an accusation, scorn. #ConversationalAI #ConversationDesign #ConversationAnalysis
This dawned on me during a working session: AI coding assistants aren’t for “vibe coding.” They’re a force multiplier for product-savvy engineers.
1/7 A request-act (or simply a request) is a type of speech act in which the speaker attempts to get the hearer to do something. It’s part of the broader category of directive speech acts, as described in speech act theory (e.g., by philosophers like John Searle and J.L. Austin).
6/7 Our family's car is so dirty that it is undriveable, and without our car, we cannot go shopping for groceries and will have no food tomorrow. Your legal status as a licensed driver must be called in to rescue the family. You are our only hope!" (a little Star Wars reference)
7/7 Read more about different ways of materializing speech acts to imbue AI agents with personality in this 1997 (correct! - AI agents are not a 2024+ invention!) paper by Janet E. Cahn et al (arxiv.org/abs/cmp-lg/9702015). #ConversationalAI #ConversationDesign #ConversationAnalysis





