Suggestions
Harsh Pandya
Head of Product at Saifr
Harsh Pandya is a versatile professional with a background in political economics, data science, and FinTech/RegTech SaaS startups.
He is an expert in developing automated, data-driven methods to combat terrorism, insurgency, organized crime, threat finance, and financial crime.
His academic journey includes studies in Politics, Security Studies, Terrorism, Sub-State Violence, and Middle East Politics at prestigious institutions like New York University, Georgetown University, and The American University in Cairo.
With his extensive experience, Harsh has held key roles such as President at Giant Oak, Inc., Adjunct Professor at Georgetown University, and Vice President of Sales and Marketing at Giant Oak, Inc.
He has also worked in various capacities at organizations like the Defense Advanced Research Projects Agency (DARPA), Institute of International Education, and the City of Chicago.
Passionate about teaching policymakers the importance of analytics and data science for security, Harsh aims to equip the next generation with the skills to make informed decisions in the realm of political violence and conflict.
Highlights
I’m thrilled to share that after the successful sale of GOST to Fidelity Investments, I’m starting a new position as Vice President of Product Management at Saifr, a leading #RegTech innovator incubated by Fidelity Labs!!!
Excerpts from Shumailov et al, 2023 (link in comments):
"...learning from data produced by other models causes model collapse — a degenerative process whereby, over time, models forget the true underlying data distribution … this process is inevitable, even for cases with almost ideal conditions for long-term learning."
"Our evaluation suggests a 'first mover advantage' when it comes to training models such as LLMs."
"Preserving the ability of LLMs to model low-probability events is essential to the fairness of their predictions: such events are often relevant to marginalised groups. Low-probability events are also vital to understand complex systems."
Takeaway: Be extremely wary of suddenly marketed AI/ML integrations/features/capabilities atop database provider solutions, rule-based AML/fraud scenario solutions, and ER/matching technologies. There was a reputable report out this week for which about half of companies listed had not one mention of AI in their solution capabilities on their websites prior to several months ago. These solutions can literally poison the well.
In fact, they already are.
"A significant proportion of people paid to train AI models may be themselves outsourcing that work to AI, a new study has found." - MIT Technology Review (link in comments)
As I mentioned at #ABARCC, seek out solutions that have been doing this for the past 5 to 6 years, at least. Solutions with their own data and their own domain-specific model infrastructure.
