ziyadb
Joined 15 years ago
Karma
692
Submissions
56
Member since
Aug 15
Recent submissions
Comment · 2 months ago
The problem of handling recurring tasks predictably comes down to the probabilistic nature of LLMs, which are based on next-token prediction.
I founded a company called Aide where our goal was to help support teams reliably deploy customer-facing agents without worrying about poor interactions. The first problem we needed to solve was making them deterministic and eliminate the variance that comes naturally with base models.
Getting them to always adhere to brand policy, eliminate hallucination, and stay grounded in data was a fun challenge. Proud to say that we’ve devised a solution that runs well and it’s worked out quite nicely in compliance-heavy and regulated environments.
Comment · 2 months ago
I was surprised to read that Anthropic (and probably other data / model companies) are doing this and it's extremely disappointing, as working towards the benefit of humanity is not an exclusive right / domain of theirs but rather, is a shared responsibility and mission carried by all of humanity itself as a collective responsibility. Thus, the preservation of this knowledge, its availability, and accessibility are the most important things that we must ensure continue.
From a historical precedent standpoint, this is akin to the burning of the library of Alexandria, where centuries of knowledge was destroyed and leaving a limited version of the history, the surviving one, and depriving successive generations of significant amount of latent knowledge.
Despite the copyright restrictions that are forcing companies to do this, they should maintain archives that are publicly available. As stated in my first paragraph, they are not the exclusive stewards of humanity despite them anointing themselves as such. Granted, a lot of these books might not be that useful, but still a relic of times pre-machine generated text, which makes them valuable if only for their archival value.
Comment · 4 years ago
shameless plug but that's exactly what we're building at Aide (https://aide.app), I would love to chat and see how we can potentially help if this is something you're interesting in learning more about or deploying!
Comment · 6 years ago
would love to learn more about how you're using Freshdesk and the types of queries you're responding to! I'm building a customer support app that uses openAI's GPT3 to generate automated responses and it's been working great so far!
How can I reach you?
Comment · 6 years ago
The holy grail [https://beepb00p.xyz/pkm-search.html#future] of this really resonated with me and fully mirrors what I've been thinking about the past few months. In my observations, it's input capture, information organization, and subsequent retrieval:
Information Capture:
Input Capture - You’re going to have all-encompassing tracking and recording of all activity, but want configurable privacy on the extent to which you want your daily conversations and observations of external things you encounter and are exposed to. Capturing input needs to be holistic and incorporate all properties of encounters and new information.
Potential sources of input:
Vision — point of view recording, see snapchat spectacles, etc as primitive examples. Audio (voice notes and multi-party conversations) - voice calls, video, etc. and other forms of audio transmission where there is more than a single party in the interaction. Digital interactions You will need to keep track of web pages you visit at what times Conversations you see on Twitter, etc.
Properties and cues must be extrapolated from the information that is captured on input, in the case of audio, transcriptions are sufficient for transcription and retrieval purposes, however since video is a visual medium, it includes significantly more properties that need to be accounted for.
The aim here is to identify sufficient data points (cues) that are subsequently represented in such a way that they are easy to search across things you have encountered but only seem to recall a certain property or cue from. This is because of the fact that human beings tend to remember things in fragments, for instance, you might remember a certain color on a page that you visited within the last 6 months and nothing else.
So long as you are capturing sufficient input and actions then you should be able to go back to any given point in time. How and where are you going to store this information? Storing everything is going to be a large amount of data. The essence of the information and context must be preserved. If you want to wind back to an arbitrary position in time with the original context intact, you want to retain as much as you can in the most efficient manner possible, so determining which data points to retain is essential. (Once the content structure has been figured out, this will be viable).
Examples of Primary Cues:
Time - humans generally keep track of things in a linear time-based fashion. Color - invokes emotion and is memorable. Physical Location - the efficiency of information retrieval is highly influenced by the location at which it is originally synthesized, encountered, and stored. Keywords - the default conventional mode. Can and should be extracted from video/imagery and audio. Imagery - search for images based on their contents and ambience.
Potential Secondary Cue — Music - see historical associated input and actions while certain music was played. (What else?)
Meta Cues — Subjects - Automated tagging of keywords/encountered content.
Any combination of these queries is possible, but ultimately the killer feature is the ability to backtrack through time to find a certain piece of information that is made available thanks to the always-on recorded nature of your interactions with the physical and digital worlds combined.
Knowing what to store, and how, + displaying it needs to be worked on further.
Comment · 7 years ago
You really need to let us try it out without having to sign up. Imagine how many users that think "oh, yet another img -> html/css tool, I wonder how questionably it works" and then get discouraged by having to take a chance on something they have no idea how well/if it works.
I'll sign up anyway and give it a shot.
Comment · 10 years ago
Interesting research, in my opinion, this been long coming as the next phase in the evolution of computing and society in general. It's how tech is going change the world to the point of 24/7 always-on connectivity; once the tools, applications, and entire ecosystem is created with more consideration to the human element of the system.
We're actually applying to YC's Summer 2016 batch with a fully automated application deployment process to accommodate for developers, by shifting focus to deployment considerations that have not been eliminated.
To whom it might interest: I've written about how this applies to deployment in some detail here: https://docs.google.com/document/d/1iU4Qez-EjHTz6o5IUVa8l18n... any additional thoughts on the subject would be appreciated.
Comment · 11 years ago
Just to add my 2 cents.
I'm based in Saudi Arabia, and considering the deserted climate and environment, there are literally no persistent water sources (aside from a few select wells that are over-exploited by bottled water manufacturers).
Desalination is currently (as of 2015) being used to supply 50% of Saudi Arabia's water needs. http://en.wikipedia.org/wiki/Water_supply_and_sanitation_in_...
So yes, it's feasible, and just FYI, it is provided free of cost.
Comment · 12 years ago
I actually found the music and the video demo excellent! Great demo of your app.
The music was especially beautiful, I even looked it up. It's Planet Love by DJ Quicksilver in case anyone else liked it.
Don't let all these comments about auto-playing audio put you down. Great experience!
Comment · 12 years ago
Sure Eric! Shoot me an email: ziyad at buildingof.com and let's do it :)
Comment · 12 years ago
After having a look and reading your interview with marketingstartups[1], it is quite an interesting idea.
It strikes me as an evolved incarnation of Buzzfeed[2] in that the entirety of its content is user-generated, whereas this wasn't the case with Buzzfeed until recently, and even now, I believe user submissions are subject to moderation.
I don't think you should give up yet. I wouldn't. I would start by simplifying the design (sure, it looks nice, but it feels over-ornamented), defining stricter guidelines for posting (e.g. encourage "Top/Best X" type posts in lieu of general free-form lists, and figure out other ways to start driving traffic to it.
I wish you guys the very best and again, don't give up yet.
PS: Please get in touch if you'd like to talk more.
[1] http://marketingstartups.com/2012/12/18/interview-with-listn... [2] http://buzzfeed.com
Comment · 12 years ago
I'm working on a blog that is intended for aspiring founders to gain some inspiration and share their stories, http://buildingof.com
Email me (in profile) if you have any ideas or would like to share your story.
Comment · 12 years ago
Hi Josh. I'm actually working on a book that is intended to provide you with the basic principles and how to actually build applications, which is something I've personally struggled with, considering the language-centricity of the majority of tutorials out there.
It's far from done, however, email me (email in profile) and I'd be happy to send you a draft.
Good luck with your journey!
Email: ziyad(at)ziyadbasheer(dot)com Twitter: https://twitter.com/ziyadbasheer