One of our fintech clients was quoted $105k USD by a consulting firm to automate their KYC with LLMs 🤯
Instead, they used
@StackAI_HQ
to build it on-prem in a week. They broke down their workflow like this:
📩 Processing customer: Receive an email with the customer documents
Just had an amazing chat with
@garrytan
and
@JTLonsdale
🔥
Can’t share much, but LLMs have immense potential to outsource business processes and unlock unreachable levels of productivity. Pumped to hear the takes from two titans of the tech industry.
Thanks for hosting us
Had an awesome conversation with Sam Altman! 🤖🚀
We discussed the future of
@OpenAI
and the evolving role of
@StackAI_HQ
as the go-to AI automation platform. The most exciting is yet to come! 🤩
Truly humbled by this chance to meet an industry pioneer.
2023 was an insane year for
@StackAI_HQ
.
📈 8 months after launch we've grown to over 38,000 users.
👥 We've built an epic team that has brought AI to 180+ organizations.
🌟 We had the honor to participate in
@ycombinator
.
Next year comes with new challenges and milestones. I
RAG vs. Structured Queries ⚔️
Want an AI to assist you with your inventory? How about reviewing your prior due diligence? Maybe analyze a large table of users?
Retrieval Augmented Generation (RAG) pipelines with Vector Databases (such as
@weaviate_io
and
@pinecone
) are great at
#FALSE
, dictatorships do not go to dialogue. This has been unsuccessfully attempted many times in the past, with
@NicolasMaduro
(and his entire party) showing no interest in improving the situation. Plus, Venezuela has never had a far-right opposition.
#GetYourFactsStraight
A US backed coup in Venezuela is not a solution to the dire issues they face. Trump's efforts to install a far right opposition will only incite violence and further destabilize the region. We must support Mexico, Uruguay & the Vatican's efforts to facilitate a peaceful dialogue.
Thanks
@svpino
!
Indeed, RAG opens an ocean of possibilities for the enterprise.
1. Sharing knowledge and training.
2. Drafting complex documents.
3. Searching over unstructured data.
4. Comparing and analyzing reports.
@StackAI_HQ
powers enterprise-grade RAG in minutes.
99% of AI applications are cool-looking demos.
Impressive, but don't get fooled by the hype. It takes a lot to build enterprise-grade products that deliver real value.
I have at least three weekly conversations with companies that want to use a Large Language Model with their
AI Assistants are the future of enterprise 🤖 🚢 🏢
Stack AI can enable you to build and export AI assistants for any division or team. You can:
🧠 Pick the best LLMs from
@Azure
,
@awscloud
,
@AnthropicAI
,
@AIatMeta
, and
@OpenAI
.
🗄️ Load documents, websites, or any knowledge
Andreessen Horowitz - Latin American Global Builders' Dinner was fantastic!
Connecting with inspiring Latinos in SF who are shaping the future is always a highlight.
Thanks to
@a16z
for hosting!
A game changer!
Open source LLMs open the door to many new opportunities:
Soon you will be able to run AI applications on premise, with your own data, without worrying about data privacy and safety.
AI is set to take the enterprise by storm.🏥🏢🏛️
Want to try it on-premise?
We believe an open approach is the right one for the development of today's Al models.
Today, we’re releasing Llama 2, the next generation of Meta’s open source Large Language Model, available for free for research & commercial use.
Details ➡️
Just had an epic reunion with
@nir_ga
and
@GalKlm
! 💪🏻
We met back during the
@ycombinator
W23 batch. Looking back just a year ago. Many experiments and hard work later, It’s been a ride to see
@StackAI_HQ
and
@traceloopdev
evolve. The best is yet to come!
Foundation models (particularly LLMs) will be the great equalizer for AI.
The $$$ days of labeling a dataset, sourcing GPUs, and training a model from scratch will soon be over.
State-of-the-art AI is in the hands of every organization.
Very soon (if not already), LLMs + RAG
Sales representatives spend over 25% of their time querying CRMs 🤯
In 3 weeks, one of our Logistics customers grew their leads per month by 36% using an AI CRM Assistant:
🔍 Receives a sales rep's question and transforms it into an SQL query.
📋 Queries their
@Salesforce
CRM.
Had a blast at the
@cerebral_valley
National Security Hackathon! Enterprise AI is here to stay 🤖🪖🛰️
Nice to run into
@realEAKlein
, PhDs are on the loose 🤓
Our very own
@BernAceituno
graced the Weights & Biases Fully Connected conference on June 7th. He stood out as one of the panelists discussing AI enterprise security among other topics - big thanks to the AWS Gen AI Accelerator Program for the invitation!
#GenAI
#AWS
🧠🤖📚
Teaching, researching and contrasting knowledge are my favorite applications for Language Models.
I have seen companies transform entire departments with a RAG pipeline. Now, this can be done even faster with
@StackAI_HQ
.
This is just the tip of the iceberg 🧊
99% of AI applications are cool-looking demos.
Don't get fooled by the hype. It takes a lot to build enterprise-grade products that deliver real value.
Every week I talk to companies that want to use an LLM with their data. The demand is huge!
Here is one idea:
The use cases
🔥Stack AI is now backed by
@ycombinator
! 🔥
We are building software to automate data labeling, detect data errors, and fine-tune Large Language Models (LLMs) for any application. Our solution streamlines ML teams' data fine-tuning process, saving thousands of hours.
🤩
@OpenAI
just announced their most powerful LLM!
We didn't skip a beat. GPT-4o is already available to users of Stack AI.
🧠 GPT-4o scores over 100 ELO points above GPT-4-turbo in the
@lmsysorg
arena.
🚄 GPT-4o is 4x faster than GPT-4-turbo.
Try it today in Stack AI
@sama
has been a source of inspiration for both myself and the countless techno-optimists who eagerly anticipated the integration of AI into our daily lives.
My respects.
🫡
i loved my time at openai. it was transformative for me personally, and hopefully the world a little bit. most of all i loved working with such talented people.
will have more to say about what’s next later.
🫡
Just blew my friend's mind by creating a Q/A bot in 20 mins using RAG, and dataset creation for LLM fine-tuning.
@LangChainAI
and
@pinecone
make life so much easier :)
code :
@svpino
It’s possible we are approaching the limits of the transformer architecture. Auto-regressive models can get us far but, ultimately, they are lossy compression algorithms…
@garrytan
@StackAI_HQ
(YC W23) has out-of-the-box support for OSS models like llama-2 and Mistral and supports multiple email boxes via
@gmail
,
@zapier
, and
@resendlabs
.
Stack AI is fully locally-hostable with our on-premise offering :)
🧪🥼 Experimenting and iterating are core to building AI products.
@StackAI_HQ
Now lets you track the performance of your AI assistants with logs. 🤩
1. Track what are the main questions from your team.
2. Find common failure modes for your project.
3. Quickly filter to find
closely monitor your flows with our latest feature… Logs!
you can now track:
- when the flow ran
- what input it was given
- what output it produced
- if it succeeded or not
…and more!
available now across all plans
The landscape of AI has shifted drastically over the last 7 days. 🤯
I recommend everybody to try Claude-3 and Mistral-Large for their projects. GPT-4 doesn’t feel like a distant leader anymore.
@MistralAI
and
@AnthropicAI
are bridging the competitive gap and they might start
🚨PRODUCT UPDATE - Integrate Perplexity for deeper insights, access OneDrive with ease, and automate with —no coding needed! Build powerful AI applications with these new integrations.
Check out the thread for more👇
#NoCode
#GenerativeAI
#AI
❄🤩
@SnowflakeDB
just released an LLM, and we're stoked about it! Arctic is already available on our platform!
We have seen a spike in applications integrating LLMs with Enterprise Data Warehouses.
🥶 Arctic delivers top-notch performance on par with models like Llama-3-70b,
One of our clients who just circulated its first AI assistants in their organization:
"I just presented the two new models to the respective teams. The response was OVERWHELMINGLY good!"
LFG 🚢🚢🚢
@gdb
AI could be years ahead if software tools were more commonplace in the ML world.
JavaScript, databases, system design, and DevOps are too disconnected from the training of ML engineers and scientists.
🚀🚀
@StackAI_HQ
just unveiled a new feature to easily compose end-to-end LLM solutions. You can connect fine-tuned or vanilla models with your knowledge base (Slack/Notion/etc), python scripts, and more!
#LLM
#GenerativeAI
🚀 Exciting News from StackAI! 🚀
We are thrilled to announce that Stack AI is now SOC 2 Type II and HIPAA compliant! 🎉
At Stack AI, the security and privacy of our customers' data are our top priorities. We are committed to maintaining the highest standards in data protection
I often wonder what is the true potential state-of-the-art LLMs if prompted correctly...
The impact of small prompt variations sometimes blows my mind. Prompting remains more art than science (almost alchemy). The gap between LLMs and AGI will only narrow once we overcome this
How do you increase a 27% accuracy score to 98%?
By adding “Here is the most relevant sentence in the context:” to your prompt.
Yesterday, Anthropic published a blog post outlining how they achieved significantly better results on the same evaluation by adding this sentence to
Chatbots 💬 > Forms 📋
Conversations are engaging and dynamic, leading to higher conversion rates than forms.
How to turn forms into chatbots? The answer is Large Language Models and Plugins. Models like GPT-4 and Claude-2 can smoothly carry conversations and interact with
Since the rumors came out about GPT-4 being a MOE model, I've been waiting for someone to adopt a similar approach.
It seems the moment of truth has finally arrived 😎
Perhaps the biggest contribution of language models in the context of knowledge management has been the ability to easily search over unstructured data.
Text embeddings with cosine similarity democratize search to any source of information.
It's exciting when a client says they love
@StackAI_HQ
. It's even more exciting when a client cites our product in an academic journal. 🤩📚
@yamanashi_univ
was one of our first partners in bringing AI to the Japanese market. We are excited to expand further.
This is even more
🔐Prioritize security, privacy and performance with Azure Open AI and Bedrock Anthropic hosted models in
@StackAI_HQ
Enjoy the freedom from global traffic with a dedicated instance for your applications. Experience enhanced performance and unparalleled security.
📣📣📣 New feature announcements
@StackAI_HQ
📨 Gmail Receiver: Now, you can easily set up your Stack AI projects to receive emails and respond automatically. Add a Gmail Receiver, and your project will be automatically executed with each new unread email in your inbox 🤯.
📩
Are AI chatbots dead? 🤔
I have found that most AI apps need flexible interfaces, especially for outsourcing business processes with varied inputs/outputs. That's why we developed the
@StackAI_HQ
form interface!
This interface can power enterprise AI workflows such as:
🏥
Don't know how to improve the quality of your RAG?
You need to know what your LLM is receiving...
📚 Announcing our chunk preview manager! Now you can see how your documents are parsed and stored in each knowledge base:
Introducing the Stack AI batch interface ⚡
A standard paradigm in AI automation is having an LLM execute a task in bulk. Now,
@StackAI_HQ
users can export their workflows to a Batch Interface, upload a CSV with thousands of rows, and automate a task on each row! 🤖 📋
🌍
📣 𝗦𝘁𝗮𝗰𝗸 𝗔𝗜 𝗪𝗲𝗯𝗶𝗻𝗮𝗿! Gen 𝗔𝗜 𝗶𝗻 𝗛𝗲𝗮𝗹𝘁𝗵𝗰𝗮𝗿𝗲 📣
We are hosting a new
@StackAI_HQ
webinar!
⏱ When: May 9th, 2024 / 10.30 AM - 11.30 AM Pacific Time
🎯 Where:
In one hour we will cover:
🤖 Understand the application of Generative
🔫 Badass! A team at the
@MistralAI
hackathon in SF trained the 7B open-source model to play DOOM, based on an ASCII representation of the current frame in the game. 🤯
@ID_AA_Carmack
A lot of people ask us at
@StackAI_HQ
: how do you properly evaluate an LLM?
Do you collect ground truth examples? Do you need humans to evaluate each input? Should you stick to simple tasks with simple accuracy scores?
One trick: use another LLM!
A🧵
🤩
@Meta
just released the most powerful Open-Source LLM on the market, and we are pumped about it!
Not only that, we just added support for Llama-3-8b and Llama-3-70b in Stack AI 🦙🦙🦙
Some quick facts:
🦙 Llama-3-70b outperforms all existing open-source LLMs.
🤖
@AnthropicAI
strikes back!
Claude-3.5-Sonnet is the newest and most powerful LLM in the market. It is also a bigger deal than it seems:
📊 outperforms gpt-4o on all most benchmarks
💸 40% cheaper than gpt-4o
📅 knowledge cutoff of April 2024
Not only that, there is an even
🕐🕐🕐
Just 5 hours after the OpenAI DevDay, we rolled out support for:
🚀 GPT-4 Turbo.
📋 JSON mode.
🖼️ GPT-4 with Vision.
🎨 Dall-E 3.
Which feature will you try first? Start stacking with our new updates today!
Summary of his speech:
1) Everyone is wrong except for me and my communist regime.
2) Venezuela is trendy on U.S. twitter and I draw tacky excel plots.
EN VIVO | El canciller de Maduro, Jorge Arreaza, en el Consejo de Seguridad de la ONU: Todo lo que han dicho aquí es mentira. La crisis en el país es culpa del bloqueo de EE. UU.
#26Ene
The position international media is playing with respect to Venezuela makes me rethink how much I "trusted" some of these sources... Using headlines referring to our president
@jguaido
as "self-proclaimed" or to
@NicolasMaduro
as "The Venezuelan leader" just makes me sick.
🎉 Exciting News! 🎉
We are thrilled to announce that Stack AI is now
#SOC2
and
#HIPAA
compliant.
🛡️ Prioritizing security and privacy is our top concern. We are committed to safely bringing AI to the enterprise.
La mayoría de los medios internacionales no explican la verdad de lo que ocurre en Venezuela.
Hoy más que nunca los Venezolanos en el exterior debemos ser embajadores locales.
Tenemos que contar la verdad de lo que ocurre en nuestro país y porqué nuestro presidente es
@jguaido
Some insights from the Jamba paper:
1. Linear scaling with context length (vs quadratic for vanilla transformer).
2. Over 3x token throughput than smaller transformer models.
3. 1/8 KV cache compared to transformer models.
Can’t wait to integrate it in
@StackAI_HQ
.
From diving deep into AI agents to forging connections with the thriving AI community, this weekend at the
@agihouse_org
was beyond amazing! Our team had a blast working on auto-generating workflows. Stay tuned for new features coming soon!
#GenerativeAI
#AGI
#StackAI
🚨PRODUCT UPDATE - Access your AWS S3 bucket with ease, use our proprietary Text-to-SQL model, and improve your RAG with the latest reranking feature.
Check out the thread for more 👇
#NoCode
#GenerativeAI
#AI
Is SQL dead? 📚 ❄ 🤖
While
@SnowflakeDB
,
@Google
BigQuery, or
@awscloud
Redshift can make structured data searchable, they remain only accessible to those who master SQL.
We have developed a Text-To-SQL LLM that lets you build AI workflows to analyze structured data in plain
Our new guide just dropped! 🙌
@StackAI_HQ
🤖 +
@SlackHQ
💬 is a powerful combination that can bring AI tools to any member of an organization.
AI for knowledge management and back-office automation at your fingertips!
New guide! 🤖 💬 📚
@SlackHQ
is the most common avenue for modern teams to coordinate and share knowledge. Now, you can use Stack AI to embed AI workflows into Slack.
Our latest guide explains how to build custom AI Slackbots and power use cases such as:
📩 Automations to