bobosha

Their lives and symptoms offer a rare window into how the brain binds experience into a single mind.

Hoover Senior Fellow Paola Sapienza and Institute for Progress’ Distinguished Immigration Counsel Amy Nice examine recently obtained data and explain why the common claim—that immigrants are hired as a source of cheap labor—doesn't hold up under scrutiny. In fact, the opposite may be true…

wouldn't LSH (Locality Sensitive Hashing) make more sense here?

We use a vector db (Qdrant) to store embeddings of images and text and built a search UI atop it.

I’m working on a new vision-language model architecture called Onida. Our aim is to match—or surpass—the performance of leading VLMs like LLavA and CogVLM, while operating at a fraction of the cost. Unlike most existing VLMs, which layer vision components onto a language model as an afterthought, Onida is designed from first principles with a truly integrated approach.

This document [1] outlines our key differentiators, and we’re now inviting beta participants to explore and test the technology.

[1] https://healthio.notion.site/Onida-Efficient-VLM-Architectur...

I largely share Yann LeCun’s perspective that scaling LLM-based approaches will eventually hit a plateau, and that a paradigm shift will be necessary. While there is ongoing debate about what that next paradigm should be, I outline my own views on the subject in this paper [1].

[1] https://www.researchgate.net/publication/381009719_Hydra_Enh...

SpaCy is awesome - we have used it in a number of enterprise-grade applications and found it to hold up well.

Nice work! will give it a try...

IMO AI is going eat most jobs - including the experienced ones - it's just a matter of time. There is no sugar coating this unfortunately.

But symbolic != hierarchical

congrats ISRO

So did everyone and their grandma also have blackberries.

exceptions maketh the rule.

5 to 10 is reportedly the magic number

>Video on the internet was not a popular thing until we had broadband internet.

I think it's an example of a post hoc fallacy. The popularity of video was in large part responsible for the investment into broadband in the first place.

how is this different from generating a feature hash of the embeddings i.e reduce from many to one embedding reduction? Could a UMAP or such technique be helpful in reducing to a single vector?

yet they do data.

This is where the memory bit comes in, if you have a memory of past embeddings and associated label(s), it could be an ANN query to fetch the most similar embeddings and infer therefrom.

has anyone tried feeding the admittedly noisy OCR-ed text -at a document level - to an LLM for making sense? Presumably some of the less capable ones should be quite affordable and accurate at scale as well.

what are some real-world usecases that people use it for?

OpenAI could very well be to this AI boom what Netscape was to the dotcom bubble. Even post dotcom crash, a lot of lasting value remained—and I believe the same will happen this time too.

I think this is a really well-designed website — great work on the look and feel! That said, IMHO, startup success is rarely something that can be accurately measured with metrics like these. There are just too many unquantifiable variables involved. But overall, fantastic job on the site itself — it’s superbly done. Kudos!

TL;DR: The human brain craves certainty and simplicity, aiming to avoid the mental effort of weighing and reconciling conflicting information.

i think this agents vs workflow is a false dichotomy. A workflow - at least as I understand it - is the atomic unit of an agent i.e an agent stitches workflow(s) together.

We used Qdrant in production - it's a solid vector db offering and highly recommend. However we are moving everything to Postgres with pgvector for simplicity i.e fewer moving parts. It was a PITA keeping data synced between pgsql <> qdrant.

The human brain operates at just 25W of power—less than the monitor you're likely using right now—whereas AI models like ChatGPT consume nearly 1GWh every 24 hours!

As I discuss in the paper, predictive coding suggests that the brain actively generates predictions and compares them to incoming sensory data (vision, hearing, etc.), prioritizing anomalies. Its efficiency stems from a hierarchical memory system that continuously updates only the "deltas"—the differences that matter. Embracing this approach could lead to a paradigm shift, enabling the development of significantly more energy-efficient AI in the future.

Ty for reading the paper! I completely agree! Assigning soft weights to the window based on context is a fascinating research area. This concept is similar to Ebbinghaus' forgetting curve, which emphasizes recency bias while requiring repeated exposure for long-term retention.

I argue that JEPA and its Energy-Based Model (EBM) framework fail to capture the deeply intertwined nature of learning and prediction in the human brain—the “yin and yang” of intelligence. Contemporary machine learning approaches remain heavily reliant on resource-intensive, front-loaded training phases. I advocate for a paradigm shift toward seamlessly integrating training and prediction, aligning with the principles of online learning.

Disclosure: I am the author of this paper.

Reference: (PDF) Hydra: Enhancing Machine Learning with a Multi-head Predictions Architecture. Available from: https://www.researchgate.net/publication/381009719_Hydra_Enh... [accessed Mar 14, 2025].

Very impressive. well done team sesame!

off-topic: i've never seen a paper with this many authors.:-)