@retttx
@hdjirdeh
@PlausibleHQ
@calendso
I’ve been collecting a landscape, so many great cos, publishing a blog post about this in next couple days…
Hadn’t heard of Jitsu and Baserow, let me know if anything else comes to mind, thanks :)
I'm back !!
After a 10 day stint climbing, fly fishing, and generally sleeping in a car around California, very happy to share that I have joined the infra team at
@a16z
, partnering with teams pushing the boundaries of AI products and research !!
@simonw
Excel… as would 99% of the world, don’t think This thread is super representative 😅
Once you hit 1M, I’d usually use pandas, but less technical folks I’ve seen use Microsoft Access… also if I wanted to plot things locally I would go to tools like Tableau public version…
0/9🚨We worked with
@replicatehq
to make
@MetaAI
‘s LLaMa-v2 immediately available to play with and embed in your apps🎁
We also took LLaMa-v2 for a spin, found examples where 13B outperforms chatGPT 3.5 (175B), and not!
More context, links and examples below 👇🧵
1/ New blog post - Open Source Challengers:
▶️ Brief recap of trends, funding & deals
▶️ Why OSS wins and why that is good for the 🌏
▶️ OSS expands into the application layer
▶️ The end of closed source software categories?
OSS Challengers Below 👇🧵
Introducing Udio, an app for music creation and sharing that allows you to generate amazing music in your favorite styles with intuitive and powerful text-prompting.
1/11
1/ LLMs are democratizing AI - data shows
@mattrickard
's JS prediction might come true, w. devs bypassing python...
Downloads of the '
@OpenAI
' lib were 10x higher in Python vs. JS through Nov '22, but are currently both growing insanely fast...
cc
@OfficialLoganK
@hwchase17
1/ Had a great time talking about OSS business models, vector databases (e.g.
@weaviate_io
), and AI alignment with
@vitalygordon
(thx for hosting!),
@chsrbrts
and
@hausdorff_space
!
Also using the opp. to share the OSS framework we discussed for the first time here on twitter :)
The ninth episode of the podcast is up!
@rajko_rad
,
@chsrbrts
, and
@hausdorff_space
talk about GTM for technical founders, Vector Databases, and what can we do about evil AI? Enjoy!
💥NEW:
@BornsteinMatt
and I made Github collection based on our LLM App Stack article!
It's a structured directory of some of the top tools folks are using to build LLM-based applications, particularly with in-context learning vs. fine-tuning (coming as well). It also has some…
Our Infra team published its initial take on the LLM App Stack back in June 2023, but this space is moving and growing incredibly fast. To help everyone keep up, and to encourage contributions, they’ve created a living list on GitHub.
Check it out here:
OSS community and engagement around
@LangChainAI
is amazing - future is bright for composable LLM apps!
📈
#1
: OSS MAUs since launch, keeping pace w. early
@huggingface
growth
📈
#2
: weekly pip installs since launch, keeping pace with
@TensorFlow
👏 👏 👏
@hwchase17
& community!
⚡️Long live the 'GPU poor' !!! ⚡️ 😅😋
But seriously, Open Source is one of my favorite values of the developer community and we are thrilled to be supporting some the individuals keeping it thriving in artificial intelligence ! 💪❤️
[New program] a16z Open Source AI Grants
Hackers & independent devs are massively important to the AI ecosystem.
We're starting a grant funding program so they can continue their work without pressure to generate financial returns.
@nathanbenaich
All great points. I think looking at revenue is quite misleading though, one has to look at margins… Open AI * could * have less revenue on every customer of Jasper, and yet end up with higher profit per customer / revenue that Jasper gains…
Awesome rundown by
@BornsteinMatt
of how we are seeing a first wave of LLM App architectures shape up! Was fun to help out and stay posted for more coming !!!
Huge thanks once again to everyone who provided input!!
New post: Emerging architectures for LLM applications
We compiled a reference stack for developers building apps on top of LLMs, focused especially on in-context learning
With
@rajko_rad
1/ Love
@mattrickard
’s writing - going to take him to bat on this one :)
I agree LLMs are much more accessible to a broader audience,
@PromptableAI
/
@LangChainAI
support for typescript proving this already - this wave is so exciting!
BUT…
@nathanbenaich
Also arguably Jasper is just the first that took off.. so long as there are 5-10 more large successful businesses, open AI will eventually make more than any of them individually, even in revenue…
For those at ICLR, we are hosting a rooftop happy hour in central Vienna on Thursday evening together with our friends
@MistralAI
(cc
@GuillaumeLample
@sophiamyang
),
@robrombach
and more in support of OSS AI!!
RSVP here - confirmations required:
Announcing Perplexity AI’s iPhone app and series A funding! Perplexity provides instant answers and cited sources on any topic, now available on iPhone. With follow-up questions, voice search, and thread history, learn and explore faster than ever before.📱
Great piece on emerging usage patterns w LLMs !
One of the key primitives are vector databases -
@weaviate_io
!
Also excited for retrieval augmentation that bypasses context with models like deepmind’s Retro (link👇)
Cc
@hwchase17
@_Brian_Raymond
🙏
Was fun to think about this with
@_Brian_Raymond
It's still early on, and things will change rapidly, but it's fun to start seeing some common patterns/abstractions emerge
More on this soon 🙂
Living in SF these days is like a perpetual science camp hangout 😅 so much curiosity, innovation and builder energy!!!
Well done
@rachel_l_woods
,
@nonmayorpete
, amazing hackathon today :)
10/ also note these bravo journos face daily intimidation and threats, including more public incidents such as
@DraganaPeco
’s apartment being broken into in 2017 (I believe 2-3 more instances of this), but also regular wiretapping and tabloid smears
I’m in awe of how
@databricks
,
@matei_zaharia
and team continuously reinvent fundamental tech as the way we use data and AI evolve over time… insanely polymath team…
Building a ChatGPT-like LLM might be easier than anyone thought. At
@Databricks
, we tuned a 2-year-old open source model to follow instructions in just 3 hours, and are open sourcing the code. We think this tech will quickly be democratized.
Excited to invite fellow ML folks w. passion for Southeast Europe (🇸🇮🇲🇰🇭🇷🇷🇸🇲🇪🇧🇦) to a happy hour 4-6pm CT, at Bryant Park NOLA
Drinks and food to be generously provided :)))
P.s. AI-provided graphic below, slightly over-indexing on the ottoman heritage perhaps 😅
If you are building with LLMs, check out and fill in
@mlopscommunity
's survey on evaluating LLMs!
➡️
There's already some great data and responses from builders. Will be an awesome community resource when it comes out!
2/ Previewing from a full report forthcoming: the power of extensibility in OSS!
Today's hottest OSS LLM? A solo developer's passion project (
@TheBlokeAI
), quantizing other models!
9 out of the top 15 players are indie: names like
@Teknium1
,
@lmsys
,
@erhartford
,
@imhaotian
!
I’m starting to think we need more public and clear data of FM/LLM economics and a timeline for when it will become sustainable for different types of use cases…
kind of like how renewable energy source costs were tracked over the years…
Potom mene pokušava da diskredituje u pismu Rektorki, dekanima, direktorima instituta, akademiji nauka... Gde me između ostalog optužuje da sam član matematičke klike odgovorne za ekonomsku krizu 2008. Pa čak pominje i potrošnju struje u kopanju bitkoina...
flox is thrilled to release the flox CLI beta to the public as an open-source project!
We're also announcing we have secured a $16.6M in Series A funding led by NEA, among others!
Ugh, how is
@stephen_wolfram
so consistently eloquent!?
Extremely well formulated and compelling vision laid out below:
Wonder if
@hwchase17
and
@LangChainAI
could get an integration going here!
Great to see✨
@metabase
✨as the most loved internal / data analysis tool in the 2022 The State of Database Survey !!
The survey includes 352 primarily full stack and backend devs and is support by
@Basedash
,
@getarctype
and
@nocodebackend
AI and Crypto are actually opposites...
Crypto:
- Tons of value capture
- No value delivery
AI:
- Tons of value delivery
- No value capture
Biggest thing they have in common is clueless people parading about them 😅
Very excited for us to partner w. the amazing
@gravicle
,
@alexyu00
, and team
@LumaLabsAI
!
More importantly, go check out the new model:
I can finally play Counter Strike as it was always intended >>> wearing traditional Serbian folk attire ofc 😇…
Incredibly excited to release Genie 1.0 and welcome
@a16z
and
@AnjneyMidha
to Luma! Also,
@baaadas
is joining us as Chief Scientist, Matt to lead applied research, and
@tuhin
is leading design.
Join us to advance intelligence through multimodal research
Per the usual, awesome work
@shishirpatil_
!! real systems-level thinking about tool-use and agentic runtimes.
Also, does
@martin_casado
have a double or something? I see him at work all day, and more often than not on calls or slack all evening too 😅
📢Excited to release GoEx⚡️a runtime for LLM-generated actions like code, API calls, and more. Featuring "post-facto validation" for assessing LLM actions after execution 🔍 Key to our approach is "undo" 🔄 and "damage confinement" abstractions to manage unintended actions &…
0/ New Year's resolution: start writing! First blog post is live - had fun with this one
Examines top trending Github repos in 2021 to better understand OSS and Dev culture
Six areas covered, each recapped below 👇🧵
1/
@MetaAI
's LLaMa v1 catalyzed a fountain spring of OSS AI innovation:
➡️Alpaca (
@stanfordnlp
)
➡️Vicuña (
@lmsysorg
)
➡️WizardLM (
@WizardLM_AI
)
OSS benefits privacy, latency, performance via fine-tuning, as well as inclusive research and community contributions to AI progress
Heavily using
@GitHubCopilot
in my work...
Small hacks: alter generated functions to control for edge cases, iterative development works really well...
It can also summarize
@ProjectJupyter
NB workflows from 10-15 cells and create a single clean function... huge timesaver...
8/ President
@avucic
’s son continues to publicly consort with top suspected mob bosses
A journalists electronics were once violently taken after making photographs of the same
@bernhardsson
Fugue is interesting. Kind of like Modin but trying to be a bit more generalized while sticking to SQL and DFs
I personally think limiting yourself to DFs might be too narrow scope down the line, generalized scale out will probs win e.g
@raydistributed
NEW: Base10 market map and research article from
@tjnahigian
&
@LOAFonseca
on Generative AI 🤖 We're seeing early signs of a platform shift that could rival that of Cloud. Full article on the Base10 website:
I ❤️ what
@suzatweet
and the
@huggingface
crew are doing with the
@BigscienceW
project in general, but just saw this live-updating tweet stream for BLOOM’s training and love it even more 😅👍
Congrats to the whole team and community!!
What happens when we train the largest vision-language model and add in robot experiences?
The result is PaLM-E 🌴🤖, a 562-billion parameter, general-purpose, embodied visual-language generalist - across robotics, vision, and language.
Website:
Kevin Scott, MSFT CTO, said this back in May btw, in response to
@eladgil
asking him about closed vs. OSS models:
"Yeah, it is an interesting thing that people are framing it as some kind of binary thing. I think you're going to have a lot of both. We still don't see any reason…
I've said it and will say it again:
#1
Smaller, cheaper, faster, more customized models will cover 99% of use-cases. You don't need a million dollar formula 1 to get to work everyday and you don't need a banking customer success chatbot to tell you the meaning of life!
#2
All…
@sama
Another interesting angle:
what happens when new gen models completely flip the system architecture / change what is possible?
prior to GPT3, companies used BERT/GPT2 to do text editing (eg Grammarly)… GPT3 made complete generation possible (eg Jasper/Copy)
What is next?? RL?
A helpful explainer from the creator of SBERT himself!
It appears research and understanding of Vector DBs and embeddings is still nascent even among leading NLP organizations... 😅
Hard problems are the best ones to solve!! Let's go
@weaviate_io
🚀🚀🚀
👨🏫Why ANN benchmarking on random embeddings is a bad idea 📊
Today
@JinaAI_
published the 1M benchmark, where some data stores like
@weaviate_io
appear to perform badly. But it is not their fault!
👎 Sadly the benchmark uses random embeddings, which is a terrible idea... 🧵
🎉 Weaviate v1.10.0 is here! Besides a lot of new features, we've released the OpenAI Weaviate module that directly integrates with
@OpenAI
's embeddings API! You can now vectorize, search through, and even mix different ML models with the power of GPT-3 all out-of-the-box!
@de_stroy
@swyx
Dynatrace, new relic and datadog all roughly the same size in revenue, though datadog growing the fastest… Splunk triple plus, elastic same size…
Most customers buy 5+ observability vendors… some of them are just the loudest / most talked about in retail investing :)
@serbia_forever
@Maja12183365
@totd_RS
Ako nema ničeg spornog sa time, zašto su i jedan i drugi lagali da se ne poznaju i da se nikad nisu videli u proteklim godinama?
Možda bolje Andreja pitaj “šta sa tim”, očigledno on zna...
@KRIKrs
2/
@LangChainAI
is one of the top frameworks for LLMs apps, initially developed in Python, JS/TS version also growing fast...
Potential separation ??
1️⃣ JS/TS is LLM frontend (prompts, chaining, agents)
2️⃣Python abstracted by langchain.llms (custom models, fine tuning)
New post: the AI Canon
We share all the papers, posts, articles, courses, and videos we've relied on to get smarter about LLMs and modern AI
Compiled by
@derrickharris
@appenz
and myself
1/ When the initial magic of LLMs wore off, the discussion turned to their limitations. But here’s why
@shangdaxu
and I are more excited about what’s coming around the bend in AI 👇👇👇
7/
- Witness testimonies in early 2000s note Serbian minister of health Zlatibor Loncar as an aid in mob killings during 90s
- photos surfaced of him hugging top mobsters in 90s
- his apartment was transferred to him by wife of top mobster
In the last month two database cos have come out of SingleStore: one OLAP, and one OLTP - how poetic!
Not sure what is says about the future of HTAP though !?? 🤔
Nice article on Vector DBs by
@VentureBeat
’s
@peterwayner
!
Vector databases let you use AI/ML to query all data as ‘vectors’ across audio, text, images and video. Unlock powerful use cases like semantic search, recommendation engines, and clustering!
Pretraining LLMs w/ Human Preferences
-Learn distribution over tokens conditional on prediction of human preference
-LLM can learn from undesirable content while guiding not to imitate it
-Better than standard practice of pretraining -> feedback
Paper
@mattrickard
@shomikghosh21
LLM Ops emerging as a layer of the stack above ML Ops tooling that support base model creation… publishing a view on this soon!
@sama
A lot of the product differentiation they build on top of an LM should hold across models, even a more intelligent model won’t necessarily be better at correctly inferring human intentions which is where UI differentiation comes from…
*LLM-DevEx*
❌It's not about the experience of devs working with your LLMs...
✅It's about the experience of LLMs working with your devtools...
But for real, this will be a thing... 1st step is good manifest files for
@LangChainAI
AIPluginTool and
@OpenAI
ChatGPT!!
Next frontier of prompt engineering imo: "AutoGPTs" . 1 GPT call is just like 1 instruction on a computer. They can be strung together into programs. Use prompt to define I/O device and tool specs, define the cognitive loop, page data in and out of context window, .run().
We're announcing the second batch of
@a16z
open source AI grants today
This cohort focuses on:
▶️ tools for LLM training/ hosting/ evals
▶️ visual AI models & communities
Thank you to the grantees for your contributions! More info in the linked post
@bernhardsson
Did 60 interviews with heads of ML/Data ~1.5 years ago while at BCG
AWS faved by software eng orgs, behind in data, investing like crazy in sagemaker, but lots of ‘vaporware’
GCP often won purely because of data offering. Tensorflow, TPUs, AI research, all huge advantages.
Welp
@jordan_segall
just saved us a bunch of time on an analysis I’ve been wanting to run myself for some time :)
Lots of investors gradient descend these metrics in our minds over time and with ‘training’, but great to have some hard data for founders to refer to! 🙏👏🙂
(1/7) Excited to introduce Segfault, a newsletter focused on devtools. First up: I collected open source traction data from 80 startups at the time of their Seed/As to create a data-driven guide to OS fundraising and what successful OS traction looks like:
🎉🎉🎉 I'm super proud to announce that we closed our Series-A for $16M, led by the amazing
@NEA
and Cortical Ventures (with the participation of
@ZettaVentures
and ING Ventures)!
just used an LM to single shot fuzzy match 50 emails with names from two sheets with different columns 😅
Idk why but somehow I'm still have a wow moment about this Entity Resolution 😅
1/ Awesome podcast w.
@j_schottenstein
,
@jthandy
and
@J_
of
@OpenLineage
Insights on bootstrapping open standards - apparently you just email
@FrankSlootman
!
Stretch historical parallels:
▶️WebAssembly
▶️Container wars
▶️Credit card networks (!?)
Thread below👇🧵
🚨🔥 new
@dbt_labs
Analytics Engineering Podcast episode is LIVE!
@jthandy
and I sit down with
@J_
from
@DatakinHQ
/
@OpenLineage
to talk about open source data standards, data lineage, and tool connectivity.
Give it a listen👇
Awesome release from
@SamiGhoche
and the team at
@forethought_ai
!!
Even cooler, you can test it out yourself on your own website and data!
Helps solve issues like hallucination and provides a chatGPT style bot with everything they need to know about your company or services!
The future of CX is here.
👋 Say hello to SupportGPT™: the World’s First Generative AI Platform for Customer Support!
Learn more about SupportGPT™, and try it out on your own data today—for free!
I’m thrilled to be joining
@davidu
,
@KTmBoyle
and the entire
@a16z
team to support founders bringing technology to mission critical industries and applications!
@bernhardsson
Not a listing of projects unfortunately, but a cool tracker for total contributors from different companies is here:
Obvs also would be cool to normalize vs. total eng org sizes...
2/ 🔊 + 🎧s ON!
I welcomed a first nephew 10 months ago, so I used
@udiomusic
to create a Balkan Brass Band song for him 🐣
NB: Turns out LMs still aren't great at Serbian (BCS) lyrics (
@gordic_aleksa
's yugoGPT was helpful!).. So enjoy these lyrics written by yours truly 🤓