Sign In

Taher Hassonjee

CEO @ DocDraft

Taher Hassonjee is a seasoned professional with a rich background in various industries. He holds a Bachelor of Arts degree in Politics from New York University. Taher has held key roles in several prominent organizations including being the Co-Founder and CEO of DocDraft, Head of Business Development at RVShare, and an Investment Associate at JumpStart Inc. He has also contributed as a Fellow at Venture For America, and as the Founder & President of the Student Political Action Club at NYU. Additionally, Taher has served as the Deputy Finance Director at Badey for Congress and interned at The Office of Congressman Joe Sestak.

Highlights

Jul 15 · linkedin

Interesting article: https://lnkd.in/ePBgRvSp

Some thoughts below:

  • Human logic remains elusive for AI. While benchmarks evolve, ARC-AGI persists as an eternal challenge. This simple test, solvable by children, confounds LLMs that ace LSATs and MCATs—proving true intelligence transcends memorization.

  • Transformer architecture excels at deductive reasoning but lacks proper learning algorithms. LLMs rely on brute force memorization, generating hallucinations that erode trust and hinder widespread adoption. ARC-AGI, a basic IQ test manageable for grade schoolers, baffles our most advanced AI. This gap exposes the limitations of our current AI approach. Despite claims of LLM intelligence, data shows their primary use in customer service with human oversight—an upgraded phone menu, far from autonomous problem-solving. Without bridging this gap, LLMs may remain call center-bound.

  • Progress demands a paradigm shift: moving beyond high-dimensional memorization to systems that infer and learn from limited data. Only then can we create AI that embodies, rather than mimics, intelligence.

Jul 11 · linkedin

SEC Chair Gary Gensler wields the Howey Test to determine what constitutes a security. His approach is of concerning flexibility—often to the SEC’s advantage and the detriment of innovative companies. Case in point: PAXOS (https://lnkd.in/eWVqzypn)

In 2022, a thriving business with over $20bn in Total Value Locked and generating $1bn+ in revenue was forced to shut down. Fast forward to 2024, and the SEC quietly drops the case—but only after the damage is done. This pattern of regulation as a detrimental guardrail, or worse, a thumb on the scale, has left deep scars in the crypto community. Now, we’re seeing a similar patterns in AI space.

The current draft of California’s proposed regulation SB 1047 seeks curtail open source and AI innovation broadly under the guise of safety. It imposes complex, resource-intensive reporting requirements on developers fine-tuning models or developing those costing over $100 million to train. The legislation’s vagueness, coupled with significant penalties for violations, creates a treacherous gray zone.

Developers are left in a quandary, unsure how to navigate this regulatory minefield without breaking the law. When governments can’t decisively establish their stance, they often shift the burden onto developers and innovators. This approach stifles progress and creates an environment where the safest bet is non-participation. As we’ve seen in crypto, and now potentially in AI, this regulatory uncertainty can have chilling effects on entire industries. It’s a stark reminder that well-intentioned but poorly executed regulation can become a significant barrier to innovation.

https://lnkd.in/eVX7d4-G

https://lnkd.in/ehfHzT7w

Taher Hassonjee
Add to my network

Location

Philadelphia, Pennsylvania, United States