Alessio Fanelli
@FanaHOVA
Followers
5,381
Following
994
Media
3,184
Statuses
28,185
Cohost @latentspacepod | Partner & CTO @decibelvc | OSS: | Writing:
Roma to San Francisco
Joined September 2009
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
GCF HEADLINER LISA
• 468281 Tweets
Georgia
• 244122 Tweets
#precure
• 80744 Tweets
Primanti Bros
• 79855 Tweets
Martinez
• 72226 Tweets
Román
• 69151 Tweets
Onyeka
• 68665 Tweets
PRETTY GIRL MANTRA
• 63577 Tweets
Bama
• 58448 Tweets
スプリンターズS
• 48393 Tweets
#仮面ライダーガヴ
• 43017 Tweets
Travis Hunter
• 40838 Tweets
#ブンブンジャー
• 38448 Tweets
Riquelme
• 34318 Tweets
Mets
• 33407 Tweets
Heisman
• 30733 Tweets
#MostRequestedLive
• 30363 Tweets
Auburn
• 26293 Tweets
Kirby
• 24228 Tweets
Milroe
• 20919 Tweets
Carson Beck
• 19149 Tweets
Saban
• 18977 Tweets
Jeremiah Smith
• 17668 Tweets
Dawgs
• 16490 Tweets
Belgrano
• 15587 Tweets
#RollTide
• 15339 Tweets
ショウマ
• 13787 Tweets
トッキュウジャー
• 11576 Tweets
いろはちゃん
• 10395 Tweets
ジンギスカン
• 10141 Tweets
メェメェ
• 10001 Tweets
Pinned Tweet
The FLOPS donation from Meta will continue until AGI is achieved 🤖
Outside of raw performance, the big change is that Llama3 allows users to fine-tune on its outputs. Founders can now fine-tune the 7B version with 405B outputs to improve on domain-specific tasks 🔥
🆕 pod with
@ThomasScialom
of
@AIatMeta
!
Llama 2, 3 & 4: Synthetic Data, RLHF, Agents on the path to Open Source AGI
shoutouts:
- Why
@ylecun
's Galactica Instruct would have solved
@giffmana
's Citations Generator
- Beyond Chinchilla-Optimal: 100x
4
16
67
1
6
20
Just making sure the APEC computer vision surveillance isn't as effective today.
16
28
412
I just open sourced smol-podcaster, your "AI intern" for podcast post-production 🐣
- Transcribes and labels speakers from your audio
- Formats the transcription for your show notes
- Creates chapters with timestamps for each topic
- Generates title ideas
12
48
314
"Oh nice, 'Transformers Math 101' sounds like a fun article"
The "101" content:
7
27
289
Working on an OSS podcast summarizer:
-
@OpenAI
Whisper for transcription
-
@LangChainAI
to summarize the content
-
@obsdmd
to store the notes
Working on using tiktoken to split transcriptions. I'm a Python newbie so all feedback welcome!
10
37
284
Why you should build RAG from scratch 🧑💻
@llama_index
is the
#1
framework for RAG pipelines with >600,000 downloads every month, and yet his creator,
@jerryjliu0
, is encouraging people to reimplement it from scratch at least once. Why?
Not enough people
4
44
277
How to get Code Interpreter to make code changes for you and commit them in git:
- Zip up your repo and upload it
- Ask it to make a specific change
- Explain the change you need to make
- Ask for a diff file that you can apply through git
Excited to explore more 👀
10
25
241
LLaMA2 is available for commercial use, as long as you don't have >700M MAU.
Products that cross that threshold:
- YouTube (~2.5B)
- WeChat (1.3B)
- TikTok (1B)
- LinkedIn (900M)
- Snap (750M)
13
17
221
WebAssembly is aliveeeee 🔥 Gemini Nano will be included in Chrome itself with WebGPU + Wasm.
5
38
217
Wtf this is already here
5
5
188
H100s? 🤨 I'll just run LLaMA2 70B in the browser, thanks!
@tqchenml
came on
@latentspacepod
to talk about the work MLC is doing to enable everyone to run models natively on any hardware / software stack, including Chrome and iPhones (and AMD cards!)
🎙️
2
40
176
Finally got my permanent green card 🙏 🇺🇸 Citizenship next!
19
3
159
Honored to have made the Forbes 30u30 🙏
I grew up in a poor part of Rome and dropped out of undergrad to start
@smarttorvy
, which failed. Neither of my parents went to college (or high school in my dad's case). I started my career in VC as a part-time engineer and didn't know
25
5
142
Say we interviewed someone about a new 405B model release. What would you want to know? 👀
36
6
114
I've been so excited by
@hwchase17
's work on
@LangChainAI
that I wanted to contribute a UI for their new project, LangChainHub.
Live @
Github:
Frontend is
@nextjs
+
@prisma
; backend is Sinatra
+ Postgres hosted on
@railway
!
7
7
108
One Year of Latent Space 🔭
Friday is
@latentspacepod
birthday, so I wrote out some thoughts on our first year of hosting it with
@swyx
. Grateful for all the support, and a special s/o to
@OfficialLoganK
for helping us kick it off as the first guest 🚀
13
13
102
If AI is so important, why is its software so bad? 🫤
@clattner_llvm
came on
@latentspacepod
to share how
@Modular_AI
will fix that 🔥
- Compilers have infinite attention spans, humans don't ⚒️ Instead of having people write CUDA kernels manually, we
6
30
102
Took us 4 months and 18 episodes of
@latentspacepod
with
@swyx
to hit
#10
on the US Tech Podcasts chart 🚀 🙏
9
12
99
.
@johnrchurch
demoing AutoPM. Assign the main Linear task to an agent and it'll create all stories and subtasks for you!
3
8
86
THIS IS NOT A DRILL
6
2
80
@LakersNation
Building up my collection.There's a couple hoodies missing since they're in the laundry lol
#LNXmasGear
http://t.co/0NK9IizQBF
5
31
69
🦙 All Llama3 release details & highlights
8B and 70B sizes: SOTA performance on most benchmarks for both Instruct and Pre-trained.
Currently training a 400B+ parameters model which will be released alongside the full Llama3 paper later on:
Created a new human evaluation set
4
18
71
Just open sourced o1 + code interpreter 🍓
- Run a prompt through o1
- Use 4o-mini + structured output to extract all code snippets
- Execute on
@e2b_dev
sandbox so it doesn't try to escape my machine
- If it fails, asks 4o to fix code and runs again
- ???
- 👽
Enjoy!
5
10
67
🆕
RAG might be a new paradigm, but indexing and search for code is something that
@beyang
and
@Steve_Yegge
have been working on for 15 years.
They just announced GA of Cody, their AI coding assistant which has SOTA Completion Acceptance Rate of 30%
2
14
65
🆕 Truly Serverless Infra for AI Engineers
In 2021
@bernhardsson
wrote a "Software Infra 2.0 wishlist", but quickly decided to take matter into his own hands by building
@modal_labs
.
On today's
@latentspacepod
we explain how VM-based clouds don't work
2
14
60
@arthurmensch
This is hilarious. We need a
@latentspacepod
ep at some point, we'll come out to France for it 😂
0
2
60
Over the last few years the team at Decibel Partners has been building the go-to destination for technical founders looking to raise capital. I'm super excited to join them to keep building on this vision, and partnering with technical founders at the earliest stages!
Big news! I am excited to welcome Alessio Fanelli (
@FanaHOVA
) as our newest partner. As an open source entrepreneur and successful seed investor, Alessio’s career embodies Decibel's “Founders-First” approach to VC. Congrats Alessio!
4
5
26
20
3
60
Is LangChain really pointless?🦜⛓️
New
@latentspacepod
episode, "The Point of LangChain", just dropped🎙️
It's been less than a year since
@hwchase17
started working on
@LangChainAI
and it quickly evolved from a side project to one of Silicon Valley's
7
10
58
Giving away a Twitch Boss Ymir skin for
@SmiteGame
, RT/fav and I'll pick a winner in a few hours.
6
80
53
Got this amazing gift from the team
@645ventures
with 130 pages of photos, stories, and milestones we accomplished in last 4.5 years. Great trip down memory lane to start the new year :) Will always be grateful for my time with this amazing team!
7
1
55
I'm always blown away by how many smart people we have in this space. Releasing our episode with
@QuentinAnthon15
tomorrow to breakdown the amazing article for us mere mortals :) Quentin,
@BlancheMinerva
and
@haileysch__
did an amazing job on this one.
0
6
54
I uploaded the slides for my "Rise of the Full Stack Employees" lightning talk at GTC on Github 🤖
TLDR:
- The abstraction of low level code syntax into semantic libraries allowed devs to go from specialized to full stack
- LLMs are semantic representations of all knowledge
4
7
50
So excited to partner with Jeremy and Eric on Answer AI 🎉 Our focus won't be on AGI, but rather on making AI useful and easily accessible to everyone, not just the GPU rich!
Today
@ericries
(creator of Lean Startup & LTSE) & I ( /Kaggle) are launching a new kind of R&D lab: . We're backed by $10m of funding from
@DecibelVC
.
For-profit R&D labs are rare today, but have an amazing history... 🧵
215
339
3K
5
11
53
The one thing more valuable than GPUs is knowing how to use them (aka make them go brrrr 🔥)
@QuentinAnthon15
of
@AiEleuther
, co-author of "Transformers Math 101", joined
@swyx
and I on
@latentspacepod
🎙️
Take your notepads out before pressing play! We
2
16
50
Chat is NOT all you need.
@thesephist
from
@NotionHQ
dropped some gems on our latest episode:
- "Conceptual backwards compatibility" in UX design.
- The workspace of the future and AI in
@NotionHQ
- The race to define the standards of AI interfaces
3
9
49
✍️LLaMA2 doesn't fit the definition of open source, but it's $3M+ of FLOPS donated to the public.
OSS has different levels of openness. AI will follow the same pattern.
I wrote about why OSS evolved like this, and how we can better label open LLM work👇
4
13
45
Introducing smol-scheduler 🐣🕐📅
Automate your meeting scheduling emails by connecting GPT-4 and your Google Calendar 🤖
You can set your general scheduling preferences in the script and then pass additional prompts in each request based on the
7
7
44
Introducing "NVIDIA Inference Microservices" 👀 Fully containerized models as "NIMs" that you can take and deploy on any NVIDIA hardware.
2
7
43
And last but not least, a big thank you to my team
@DecibelVC
who helped with hosting, and to my friend and co-host
@swyx
! We went from an idea for a 20 people event 2 weeks ago to to hundreds of people showing up yesterday 🙏 More Latent Space events (and the podcast 👀) soon 🔭
1
1
44
🎙️
@latentspacepod
with
@tri_dao
, author of FlashAttention, is out ⚡️
- Technical deep dive into FlashAttention, how it works, and how Tri got the inspiration for it
- FlashAttention2,
@nvidia
CUTLASS 3, and the Hardware Lottery (s/o
@sarahookr
)
-
2
12
43
2
0
40
We had over 130 people come tonight to our AI World's Fair 🙏 Can't wait to do this again with even more amazing demos!
2
2
37
We have 360+ builders who RSVPd for tomorrow's demo day 😱 It will be the biggest AI demo day SF has seen yet! RSVPs are now waitlisted, but even if you're approved, we can't guarantee everyone will be able to get in, so arrive early! Agenda: (1/2)
1
6
38
First firm to have an office
@poapxyz
?
The next 1000 visitors to our
@645ventures
San Francisco office can mint a unique POAP 😱
15
5
34
Our episode with
@soumithchintala
is coming out this week on
@latentspacepod
! Two chapters preview:
- [00:38:42] How do you allocate 600,000 GPUs?
- [00:42:05] Even the GPU Rich are GPU Poor
😅 One of our best episodes yet imo
2
6
36
So excited to be promoted to Principal on our Investment & Research team 🎉 When I moved from building our Voyager platform to investing I had a lot to learn and I'm thankful to have had
@AHolidayiii
and
@thedatavc
as mentors over the last 4 years. I learned a lot from you guys!
7
1
35
We got sooo much interest for this that we're turning demo day into an "AI Open House" on Feb 17th afternoon.
We'll have two floors of coworking space, project demos, prizes, food, and more! 🙌
RSVP:
2
4
35
Working on smol-scheduler (out soon!) and GPT-4 (right) blows Claude2 (left) out of the water here:
- Claude2 doesn't seem to be able to distinguish between workday and weekend, while GPT-4 can.
- When prompted to show times for the next 2 weeks, Claude2 gives you times in the
6
1
35
AIUX 2024 is live 🎉 500+ signups and ~50 demos scheduled!
@swyx
@thesephist
@Mappletons
@geoffreylitt
6
2
34
Man this is heartbreaking 💔 RIP to one of the realest.
I stayed at your place in LA when I interned, and you let me use one of your laptops because I didn't own one 😅
Being a part of RG easily changed the trajectory of my life, both professionally as I first learned about
RIP Mahbod, founder of RapGenius. You were a legend in your own time.
167
1K
8K
4
0
33
One day you're building a prompt battler at the
@latentspacepod
hackathon, 1.5 years later you get acquired by NVIDIA 🫡 Congrats Nader 🔥
Exciting news...
🚨 has been acquired by NVIDIA! ❤️🤙
We started Brev with the goal of building the best damn developer experience possible.
We’ve been working closely with NVIDIA since August, and we couldn't be more excited to team up.
I'm
261
90
1K
1
4
32
This SVB situation is just another example of the doomerism that has taken over recently. Many just look for a catastrophe narrative and amplify it to get likes. Feel bad for my friends at SVB, but especially for all the founders put in a bs situation because of hysteria.
1
4
32
There are sooooo many OSS AI tools, agents frameworks, etc, that look like this:
- Launched in fall of last year
- Very fast initial stars growth
- No commits in months after initial hype
The drop off rate is extremely high.
10
3
30
We hosted our first
@latentspacepod
x
@DecibelVC
AI Pioneers Summit yesterday, the largest get together of top AI founders and enterprise leaders 🤖
Here's a recap of the amazing panels, and of our newest Latent Space announcement 👀
2
7
31
I spent 2 hours talking with Emmett at a dinner a few weeks ago, this was NOT on my bingo card (and probably not on his either lol)
3
0
32
Broke: RLHF
Woke: RLHB(ehavior)
@swyx
and I had a great conversation with
@paladin314159
and
@isnit0
at the
@Amplitude_HQ
Product Hackathon 💻
Amplitude generated $230M ARR in '22 by helping users capture product usage data. We talked about how crucial that data is in the
1
4
30
🤖 If you're also building products powered by AI, you've seen first hand how unruly they can be.
@ShreyaR
is building
@guardrails_ai
to solve that:
- Enforce structure of outputs
- Validate correctness with "contracts"
- ReAsk and correct if needed
1
4
30
Latent Space Infra panel w/
@swyx
@ClementDelangue
@jefrankle
@SethuramanPriya
@vipulved
🔥
(Thanks
@Lux_Capital
for hosting!)
0
8
29
The new home of
@latentspacepod
🚀
INTRODUCING SOLARIS AI
A 20,000 sqft office building for AI startups
Featuring:
- 16 offices
- 70 desks
- a floor for focused work
- a floor for serendipity
+ more
Get deep work done. Be surrounded by the most talented builders in the city.
Apply:
109
43
659
6
1
29
This is the single biggest issue with
@cursor_ai
Composer / `Apply`.
When the model gets lazy and puts a placeholder where your existing code is and then you Apply, it will delete the actual code that already exists. If anyone has a fix other than prompting "Don't get lazy" lmk
6
0
30
We have a new entrant in parallel calling 🫡 Updated slide if you attended the Latent Space function calling workshop I hosted a couple weeks ago.
This is one of the most underrated features of LLMs.
@PSH_Lewis
@FanaHOVA
@aidangomez
I think that’s referring to the model making multiple function-calls in one hop (whether that’s the same function with different params or just different functions). And yes command-r does that off-the-shelf!
@FanaHOVA
Welcome to test it using our tool-use API!
0
0
5
6
2
28
What are the odds that NVIDIA gets nationalized? 🤔
That Q was enough to get
@realGeorgeHotz
to start tinycorp with one goal: commoditizing the petaflop.
He came on
@latentspacepod
to tell
@swyx
and I the story. If AI is the most important tech shift of our time, access to
2
9
29
.
@aiDotEngineer
is on 🔥 Got to meet so many
@latentspacepod
listeners, please do say hi if you see me tomorrow / Tuesday 🙏
1
2
28
How AI is eating Finance 📈
@vagabondjack
is back on
@latentspacepod
! He shared all the AI Engineering wisdom he acquired while turning LLMs into AI thought partners
@brightwaveio
for customers with >$120B under management 💰
- Why he lost faith in long context windows
- 3
5
8
29
Sarah and I just moved to San Francisco! Excited to explore the city, and meet many of my long distance friends in person more often :) You know where to reach me.
4
0
27
Our first guest is
@OfficialLoganK
, the first dev advocate at
@OpenAI
. We chatted about ChatGPT, GPT-4, how to build LLM apps, and more 🔥 You can listen on Substack, Spotify, Apple Podcasts, etc!
3
3
27
In 2017 I applied to work as an engineering fellow
@645ventures
, then an $8M fund working out of a windowless room at WeWork on 401 Park Ave. Today we are announcing our $160M Fund III, which we will use to back exceptional Seed and Series A companies 🙌
5
1
27
I used to be a YouTube creator 10+ years ago and graphics were always such a pain. Creating thumbnails / posters with DALL-E 3 is like cheating (it does have some typos though lol)
4
0
27