Something gives me a feeling that GPTs will be dead in sometime like plugins.
A general purpose RAG solution is only good to wow people, but the more you use it, the more flaws you discover.
As a CTO at
@aidetic
, having hired over 50 Data Scientists, I've got some valuable insights to share about acing those DS interviews.
Let's dive in! 👇
@C_P_Gurnani
We took Sama’s comment to heart and launched our direct customer facing application.
The base framework
@getconvflow
is for enterprises to build their private self hosted chatgpt like systems.
is a direct customer facing application built on top of
So people talking about
@OpenAI
killed a bunch of startups.
@sama
released couple of wrappers and people going berserk.
In my humble opinion, they did a great great job with foundation models, but applications side not so much.
Their RAG, is not the best out there and it’s
@mustafasuleymn
For people asking what are few ways to solve hallucinations:
1. Controlled generation: Implementing techniques to guide or control the output of the model can be helpful. For example, using prompt engineering or specifying constraints in the generation process.
2. Ensemble
@VaibhavSisinty
This would affect lots of agencies and vendors in India.
Loosely speaking smaller agencies have 1-2 months of runway. Companies in US will have to priorities who to pay, who not to pay, when to pay.
@codyschneiderxx
We work with multiple such clients, where 3-4 engineers from our firm, at $15-16K per month, help startups build/maintain their complete product.
Some of those startups have crossed Million $ ARR. Some have closed multiple enterprise clients.
We have a cross functional team
In the last 6 years, we have successfully implemented AI in industries like sports, media, bfsi, accounting, retail, manufacturing etc
We accepted the LLM challenge with
@getconvflow
a couple of months ago.
The idea is to build domain specific LLMs for enterprise and provide
@paraschopra
Looking back, I used to believe that perfection and optimization were achievable on the first try. Now I know that iteration and improvement are essential for progress and growth.
We took Sama’s comment to heart and launched our direct customer facing application.
The base framework
@getconvflow
is for enterprises to build their private self hosted chatgpt like systems. is a direct consumer facing application built on top of
OpenAI founder Sam Altman said it’s pretty hopeless for Indian companies to try and compete with them.
Dear
@sama
, From one CEO to another..
CHALLENGE ACCEPTED.
Landed our first paying customer
@fabrichq_ai
today, after a hard chase of 2 months. Perseverance pays off.
Will keep moving forward with the same passion and vigour, building the best AI automation tool out there.
First in our journey, many more to come.
Super psyched for the
@sherjilozair
Agreed. So many research directions are there to mention a few
1. Making transformers efficient to reduce computational complexity and improve memory efficiency.
2. Hybrid models: Combination of transformers with other architectures, such as convolutional neural networks (CNNs)
Looking for
@bubble
developers for some paid consulting and development work.
Need to upgrade a website and add more features.
Please follow and dm your hourly pricing and past portfolio.
#bubble
#nocode
#softwaredevelopment
@chandrarsrikant
@svembu
@deepsekharc
In the last 6 years, we have successfully implemented AI industries like sports, media, bfsi, accounting, retail, manufacturing etc
We accepted the LLM challenge with
@getconvflow
a couple of months ago.
The idea is to build domain specific LLMs for enterprise and provide them
“It may just be a wrapper, but it’s a very small problem, openAI won’t build that”.
Now that openAI has launched chat with PDFs and documents. What’s going to happen to all other chat with pdf tools?
One good thing was that a lot of them were charging one time fee for
📣 Calling all fine-tuning enthusiasts! Let's dive into the world of LLM fine-tuning methodologies together. 🚀
What specific fine-tuning techniques or strategies have you found most effective when fine-tuning open source language models like LLAMA, Falcon or similar models for
“
@OpenAI
’s Custom GPTs are lame and overrated and not at all useful.”
Tweets like these were flocking on the timeline, so I decided to run a small experiment, to get into the depth and understand the reasons behind these tweets and reasoning.
Here are my thoughts:
First
A staggering 80% of AI/ML projects don't make it to production? 🤯
It's a sobering statistic, and having overseen 150+ successful AI projects as the CTO of an AI consulting firm over the past 5 years, I've seen some common pitfalls.
Let's explore the top mistakes and how to
Infosys in their Q3-FY24 press conference said “they have 1 lakh employees trained in generative AI”.
In my experience of interviewing a lot of
@Infosys
developers with 2-3 yrs experience:
1. Most of them don’t even know git.
2. Have worked with just 2-3 python libraries. I am
Seeing people talk about Sama's comment all over twitter, but hardly any action.
We have been working and experimenting with fine-tuned self LLMs for last couple of months.
We launched a user facing app built with
@getconvflow
.
Try it out here:
As a CTO at
@aidetic_in
, I have interviewed more than 500 candidates for DS roles, and have been interviewing for last 5 years before AI was cool.
I have seen a lot of people making these mistakes.
Here's what you can do to avoid them and make a lasting impact on the
@TechBroDrip
Sam To ModiJi:
If you’re not using ChatGPT someone using it will replace you. Here are 10 prompts that can take your productivity out of the world 🚀
@malpani
Well, passing a tough exam and recruitment process once in their life makes IAS officers fit for ruling us for the rest of their life because... they must have mastered the art of sitting in a chair for extended periods of time without losing their cool!
@rakyll
Generative AI has democratized our ability to communicate with machines on a whole new level and getting complex things done with a level of ease. It feels like the power of high-level programming languages amplified.
@VaibhavSisinty
Companies will prioritise paying employees before the vendors, Which means delays in payments.
Agencies dont have investors, who can pool in money in case of emergency.
Which mean layoffs.
Just talked to a data scientist with 1 year experience from an NIT.
Was working for a big E-Commerce company. Resigned 2 months ago.
Started looking for new jobs, without any luck.
Starts a small side project, creates a Github repository. Changes LinkedIn title to Founder and
@mohakmangal
Anyone building chatgpt response reviewer?
Which basically integrates in input fields (emails/social media/etc) and looks for such errors.
@minchoi
@jerryjliu0
@ivanmartit
Building something very similar with
We strongly believe that opensource LLMs finetuned on private data and hosted securely on enterprise VPCs are the way forward!
@sabakarimm
Provide apis for companies to build and scale their own private and customised chat gpt.
Website is not yet ready, Planning to launch by end of this month.
SAAS looking to integrate content search, information extraction and conversational interface in their application.
@amasad
Chaining: Ensemble of models, probably general to specific, for more accuracy.
Retrieval: Extracting relevant information from various information sources mostly text
Agent: Building applications to do tasks on your behalf.
Hasn’t these been the goals for language related
No one gave us 10M$, let alone 100M.
Still we went ahead with building what we think will be the future.
@getconvflow
provide domain specific language models and surrounding infra for LLM related operations like training, fine-tuning, testing, RLHF, deployment, monitoring etc.
To all those who are asking.. Yes, the CHALLENGE IS ON!
The purpose of this challenge is not one-upmanship or competition or headline hunting.
It is a challenge to ourselves -- to push the frontiers of technology for nation building.
I’ve always believed that AI is not
@sesigl
- Over engineering
- neglecting user experience
- lack of modularity
- lack of testing
- poor error handling
- inadequate error reporting
- poor documentation
- ignoring security
And the list goes on.
Controversial Take:
Recursively calling GPT-4 like AutoGPT does is a mistake. The AI errors and hallucinations just compound exponentially. No practical use for now, only over-hyped tweets from influencers desperate for followers.
Elevate your
#SEO
game effortlessly!
I've crafted an automation that suggests SEO keywords and creates a 1-month
#content
calendar.
For now, it's for personal use, but if this gets 100 likes, I’m willing to put in some work to make it public.
Cant do this with
#chatgpt
@VicVijayakumar
The unfortunate situation with SVB is that it's not the affluent individuals who’ll suffer the dire consequences
Instead, it's the numerous small companies that were obliged to maintain their funds in SVB. And even smaller companies in business with these companies. Deeply sad.
What if I say we have better RAG than
@OpenAI
?
Over last 2 months, we have been working hard to build this no-code AI automation tool
@fabrichq_ai
. The idea was to give AI capabilities in the hands of business users.
We made
@fabrichq_ai
flexible enough to handle various RAG
This financial year, my AI Consulting business has achieved a monumental feat - we have skyrocketed our revenue by a whopping 129%, crossing the coveted $1 million mark! 💰
@vaibhavbetter
I dont agree.
We at aidetic are a team of data scientists working on cutting edge AI for last 5 years. All young , driven, passionate developers. None of us is PhD.
Have worked with multiple data formats cv, nlp, speech, structured. Multiple industries retail, sports, real
AGI has been achieved internally.
@sama
commented after 7 years on Reddit and this is why he has to say.
What’s going on?
Will OAI announce AGI on Nov 6?
@marktenenholtz
I run an AI consulting firm, focused primarily on building AI powered apps. We have been in business for last 6 years, worked with multiple startups across industries and geographies.
One common pattern that i see is people starting with deep insights about the problem are the
Reverse countdown to AGI?
Dealing with images is one thing, but the thought of models recognizing different chart types and grasping their intricacies is quite amusing!
Blows my mind off!
Imagine it understanding workflows and actually implementing them. Which is kind of what
@vaibhavbetter
We have grown our startup to 1M$ ARR without any external funding.
Always operated with limited financial resources, with a mindset of extreme frugality and resilience, finding innovative ways to stretch our resources and make every penny count.
One thing that we have learned
From chatbots to sentiment analysis, natural language processing (NLP) is everywhere! With
@LangChainAI
, a powerful Python library for NLP, you can unlock the potential of this technology and explore the possibilities of text data processing.
In the last 6 months, I’ve used X to close north of $50K in sales for my AI consulting firm.
How?
I built a social listening AI tool using Fabric that uses the X API and GPT-3.5 to give me 10-15 high intent leads every month.
If this gets 500 likes, I’ll open it up for anyone
I think
@sama
and
@OpenAI
should work on making sure more and more systems go to production using their foundation models and not focus on building consumer apps and marketplace and app infrastructure around it.
Make the models faster, better and cheaper.
@OpenAI
is making them
Want a free Custom GPT trained on
@naval
’s wisdom?
Will ingest his books, podcasts etc etc. No need to pay 20$ subscription.
200 likes and the bot is yours.
Show some love.
If you want free custom GPTs for anyone else like
@elon
,
@sama
@paulg
etc. Feel free to drop that in
I can resonate very well with what
@yoheinakajima
is saying. We
@fabrichq_ai
are working on somewhat similar vision, but not limited to VCs.
Every business, every function needs be tech enabled. AI and automation will be a major driving force in making that a reality.
A16Z hit the market saying “VCs don’t add value”, built a massive portfolio support back-end changed the VC landscape and made “value-add” standard.
At Untapped, we’re saying “VCs are lazy and inefficient”, and AI and automation can solve that. We see large firms increasingly
We recently got rejected by
@Accel
Atoms program.
Some of the criteria’s that they mentioned were competitive intensity, market size, distribution or technical moat, founding team’s experience, gtm etc.
We have been doing AI consulting for last 5 years, before it was such a
@yoheinakajima
- How do I parse and save tables or charts?
- How do i tag documents with meta data? If same information is present in multiple documents how do I know which one is right? For ex quaterly reports.
- What’s the right chunk size?
- Do i take a last few exchanges and rewrite query
Hearing a lot of people are facing issues with the OpenAI APIs. That's what our customers at Fabric are telling us as well.
Must be super hard for the OpenAI team to ensure API uptime considering what they've gone through in the last 2 days.
But Fabric already has a native
@debarghya_das
This looks more like a human behaviour issue, rather than rational financial issue.
Selling 10yrs low yield assets for 3yrs high yield assets would increase net interest margins.
And they have good liquidity on books. If vcs and startups dont go super crazy about withdrawing
Even
@OpenAI
didn’t launch for the scale.
Lesson for early stage founders here:
Don’t build for scale on day 1.
Maybe leave provision for 1 next update and be mindful of potential bottlenecks.
If you have to scale that’s a good problem to have, but reaching out to
usage of our new features from devday is far outpacing our expectations.
we were planning to go live with GPTs for all subscribers monday but still haven’t been able to. we are hoping to soon.
there will likely be service instability in the short term due to load. sorry :/
Have you heard about
@FastAPI
? It's a modern, fast (hence the name), and easy-to-use Python web framework for building APIs in your Data Science project.
If you haven't tried it yet, you're missing out on a great developer experience.
@mehran__jalali
We are building
@getconvflow
for enterprises to be able to do so.
It’s a framework which provides infrastructure and tools to support LLM related operations like data preparation, fine-tuning, RL, testing, deployment, versioning and monitoring etc.
Basically it allows
All must have seen those interesting insights posted by
@deepigoyal
on the NYE, which drove huge engagement and turned a lot of eyeballs.
Read on to find out wha't under the hood.
How
@zomato
Supercharged Its Data Analytics with AWS Graviton?
🧵
I am slowly taking on the Product role
@fabrichq_ai
, I come from the a Tech background.
This is my first lesson, as a product manager:
“Can’t afford to let product design be an afterthought.”
@seanjtaylor
Not just AI chatbots, vector dbs could also be used for:
1. Visual search
2. Problems like face recognition
3. Recommendation systems etc.
Basically for searching and matching problems.
Our company, Aidetic, has been featured in ET Tech by
@EconomicTimes
and it's a proud moment for our team!
India will become a powerhouse for AI Consulting. And we're excited to be contributing to this mission 5 years and counting.
I'll share a full blog written by our team of experts on how Startups and Small -Business Enterprises can deal with 12-14 hours of server outage and save themselves from going bonkers.
Please feel free to engage, I'll reach out personally.
#instagramdown
#Instagram