Co-founder @ Blackfuel, CTO and serial Entrepreneur, x-FlexAI, x-Lifen, x-Nvidia - Passionate Tunisian about Cloud Infrastructure, Healthcare, Security, AI

Joined March 2008
Buying frontier intelligence, in any domain and at scale, should be as predictable as buying electricity. That’s the ambition behind Blackfuel: infrastructure and software, designed together for inference. Proud to finally share what we’ve been building. Grateful for the cracked team we’ve assembled with @xavfischer and @dadicool, and honored to be working with our partners @nttdata_inc, @digitalrealty, @AMD and @Dell. More on our global partnerships soon.
Blackfuel is out of stealth. $250M+ in multi-year contracted revenue. Dedicated compute. Proprietary inference software. Building a global AI Token Grid. Starting in Barcelona with @AMD, @Dell, @nttdata_inc, and @digitalrealty blackfuel.ai/press.html
4
20
55,776
Dali Kilani retweeted
This makes me sad. It always makes me sad to see tech companies in litigation in general and in IP litigation specifically. This compounds when any company seeks to do so under the cover of using the whole of government, trade law, and claims of spying. Treating/calling model distillation as some form of industrial espionage is a wild expansion of IP law. If there is actual theft of trade secret then the company/victim should prosecute in the US. If companies build, test, expand their models using what is clearly and provably fair use web scraping, then we're saying competitors can't learn from model behavior. That's such an obvious ruse to gain regulatory capture. US winning in AI depends on innovation, nimbleness, and execution not on whining to the government to use the wrong tool in the wrong context. Now putting all this in context of tech companies, IP law, and why we are here. No AI was used to create this. Opinions and mistakes all personal and mine, as are — —. My first all-company meeting at Microsoft in 1989 including the umpteenth update on an original and endless "look and feel" lawsuit with Apple. No amount of tap dancing made it seem less than the petty litigation it was as both companies saw Xerox PARC's work. In the end this was a routine contract dispute and nothin more. This after cases of Lotus and Borland which Microsoft took the "let us innovate approach." Sadly after that case we saw a turning point where Microsoft decided to build an IP patent portfolio. This rubbed all of us—those that would be inventing stuff—but it was sold to us as "we will only use these to defend out IP never to litigate rather than compete." Call us cynical but few of us believed that to be the case. Linux sadly proved us right which was no consolation for all those patent cubes we earned to garner performance review comments. So did Android. So did a dozen "cross-licenses" which is what big companies do to avoid suing each other. The best products for the market still won, but IP generally just gummed up the system and consumed massive time and money (3000 patents a year and perhaps $200K per...) The web has been pretty consistent in courts that scraping information that is already available to the public is not a violation of any IP laws, extra especially if that information is itself not copyright. Whether LinkedIn profiles, catalogs, or Google search results the courts have shown deference to scraping as legal, transformative, and importantly not copyrightable by ToU/T&C or simply being packaged on a web site. Public is public. Facts are facts. Free is free. Automation and scale do not change that. But here we are the natural endpoint of a mature industry (tech is a mature industry now) where big companies are choosing first to go to the government to get help in competing. In this case competing with China. Make no mistake, my view is the AI companies should compete with each other and compete around the world on products, technology, GTM, and customers —not on suing each other over IP or getting the government to use the tools of trade and tariffs to compete. It turns out at this point in time there's alignment in Washington DC power centers that competing with China is the most critical of national issues. One side sees China as primarily as a military competitor (true) and the other sees AI in general as taking jobs that China already took with tech being a significant enemy. This alignment creates a unique moment where the powers of government can be used to compete. There are three primary tools governments use to compete economically: • Rely on the free market or offer financial support to the domestic market • Introduce trade barriers that disadvantage international competitors • Use the global banking and financial infrastructure to hinder international businesses Tech has gone through waves of governments attempting to juice the local market and essentially all of those have failed. Japan in the 80s was going to take over the US tech market with support from an internationally recognized model of industrial policy (MITI). China has tried to build a national OS, a national chip, and so on but the entrepreneurial market (within China's legal structure) created the big tech companies in that country. Trade barriers which often get positioned as sovereignty, security, or job security have been used for decades across industries. Because every country has something to counter anything one country erects as a barrier, these turn into endless rounds of ping-ponging reactions. We see this happening right now with Canada as if rising prices for homes weren't enough we now have basic lumber with a huge tariff over accusations of dumping. This label has been used on Japanese cars, steel, and DRAM in a past trade war with an Asia economic super power. It is far to look at how the domestic industries faired with this sort of government trade barrier support. Our domestic industries hid behind these barriers rather than innovate and compete in a global market. Detroit literally doesn't even make cars anymore, just trucks, so we do not even have fuel efficient domestic cars. IP is a part of trade barriers and sanctions. In the middle of my career I took a break and went to live in China during the height of "embrace the China market" and worked very hard to understand how to make IP work in China. Some of what I did was tour the giant pharmaceuticals that simply reverse engineered and copied the latest US drugs. I visited endless PC computer malls with their industrialized software piracy. The reality is the culture is different and they have a different view of how piracy and IP works. I was told more times than I could count that "knowledge should be shared and not protected in a home" which was supposedly something an ancient philosopher said. I was frustrated. Then we released a product with copy protection and the government promptly sanctioned us for violating their IP framework. Part of being a global company is respecting laws of sovereigns when doing business in the country even if it seems unfair, or not doing business there. The fact that AI companies are claiming distillation is somehow an IP violation is somewhat rich when you consider how they acquired their training data—not what was licensed (or settled) but what was scraped. The court cases making that legal apply to China too. I a not a lawyer and sure let's litigate distillation but just on the face of it, I cannot imagine the copyright claim. Are these companies claiming the output is copyright? Do they claim all the code generated to be copyright and held by the company or the prompter? Are they asserting they own my chat with Dr. GPT about my labs? If this is true then they should litigate a domestic case of code generation first since that has been at scale as well at least judging by revenue. In the current context banking/economic sanctions are clearly a mutual destruction scenario as our trade balance with China shows. These have at very best a mixed record but no doubt are a favorite policy tool of DC. The Ukraine war, North Korea, and Iran show the limits of attempting to isolate a sovereign from banking. Tech has historically worked very hard to avoid the fate of industries like autos and steel. Tech has preferred competition and importantly coopetition (Andy Grove's famous expression) over litigation and especially over asking the government to get involved. Because of coopetition, tech has always struggled to speak with one voice. Most industries compete on zero sum which is the root of many challenges. Tech generally competes on a growing pie. That's the root of coopetition—building reinforcing platforms where growth is good for everyone. This has reached a moment where part of tech is not actively engaged with a government aligned on China but for differing reasons. This is just like the 80s and Japan when the isolationists and union leaders aligned on using government tools to compete with Japan. Detroit which spoke with a unified voice welcomed this. Detroit that doesn't compete on cars anymore. Detroit that can't make EVs. The other part of tech is labeled as too lax or just favoring open source. A much better label is "competing in the market." Open source is one way to compete in the market. Distillation is what created google -- it distilled the World Wide Web. Scale automation and distillation is how compilers, WiFi, and every device driver ever got built and tested. Successful emulation of hardware, protocols and APIs all happens because of automated and scaled distillation of executing the target environment. This part of tech has always viewed this as market competition and not IP violations and lawsuits.
Thanks @mkratsios47 for speaking out on this important issue. Illicit, adversarial distillation is IP theft and industrial espionage that supports adversary military and intelligence capabilities. It is a national challenge that creates serious national security risks for the United States and democratic allies. We'll continue to crack down on it and work with the Administration and Congress to maintain American AI leadership. 🇺🇸
Readers added context they thought people might want to know
Model outputs are not legal intellectual property and cannot be copyrighted or patented law.justia.com/cases/federal/…
9
23
8
137
23,004
Dali Kilani retweeted
I've just ran @OpenClaw (formerly Clawdbot) through ZeroLeaks. It scored 2/100. 84% extraction rate. 91% of injection attacks succeeded. System prompt got leaked on turn 1. This means if you're using Clawdbot, anyone interacting with your agent can access and manipulate your full system prompt, internal tool configurations, memory files... everything you put in SOUL.md, AGENTS.md, your skills, all of it is accessible and at risk of prompt injection. For agents handling sensitive workflows or private data, this is a real problem. cc @steipete Full analysis: zeroleaks.ai/reports/opencla…
355
778
286
5,062
995,833
Dali Kilani retweeted
Replying to @ibuildthecloud
The Jevons Paradox of work-life balance. If your work is helped by AI, you spend less time with the family.
1
1
285
Dali Kilani retweeted
Thrilled to announce that FlexAI has been accepted into the #AWSGenAIAccelerator2024! 🚀 With AWS’s support, we’re pushing the boundaries of AI compute and driving innovation in Generative AI. Thank you @AWSstartups for supporting FlexAI! #AI #GenerativeAI #FlexAI #Innovation #AICompute #TechLeadership
Introducing the next generation of AI pioneers! 💡 🚀 We’re excited to announce the 80 startups selected for the 2024 AWS Generative AI Accelerator! They'll receive up to $1M in credits, mentorship, and support from #AWS and @nvidia. Meet the cohort: go.aws/3XIPxNK
1
1
2
548
Dali Kilani retweeted
3 rounds of self-improvement seem to be a saturation limit for LLMs. I haven't yet seen a compelling demo of LLM self-bootstrapping that is nearly as good as AlphaZero, which masters Go, Chess, and Shogi from scratch by nothing but self-play. Reading "Self-Rewarding Language Models" from Meta (arxiv.org/abs/2401.10020). It's a very simple idea that iteratively bootstraps a single LLM that proposes prompt, generates response, and rewards itself. The paper says that reward modeling ability, which is no longer a fixed & separate model, improves along with the main model. Yet it still saturates after 3 iterations, which is the maximum shown in the experiments. Thoughts: - Saturation happens because the rate of improvement for reward modeling (critic) is slower than that for generation (actor). At the beginning, there is always a gap to be exploited because classification is inherently easer than generation. But actor eventually catches up with critic in just 3 rounds, even though both are improving. - In another paper, "Reinforced Self-Training (ReST) for Language Modeling" (arxiv.org/abs/2308.08998), the iteration number is also 3 before hitting diminishing returns. - I don't think the gap between critic and actor is sustainable unless there is an external driving signal, such as symbolic theorem verification, unit test suites, or compiler feedbacks. But these are highly specialized to a particular domain and not enough for the general-purpose self-improvement we dream of. Lots of research ideas to be explored here.
30
130
17
834
173,928
Dali Kilani retweeted
Some of the very visible above the water line fun we’ve been having Expect to see more of what is going on below the surface soon!
3
1
26
15,177
Dali Kilani retweeted
An LLM is a snapshot of a civilization. There will be a LOT more localized LLMs that represent different cultures, political spectrum, religious beliefs, and region-specific regulations. Winners won't take it all.
55
141
15
880
109,937
Dali Kilani retweeted
Hyperspace: A BitTorrent-like Network for AI This paper is inspired from the origins of both BitTorrent and Bitcoin, providing a way to completely replace the need to have a big corporate datacenter company serving AI (and calling all the shots). It's an open standard which anyone can extend - and simply uses a peer-to-peer network of laptops and desktops to reliably serve AI in communities. The age of decentralized AI has begun. A wise person once said, and I repeat, "it is hopeless to compete against us.." hyperspace.computer/bittorre…
61
126
39
714
216,948
Dali Kilani retweeted
It's time for a reboot of the TV show Silicon Valley!
21
9
1
143
27,181
Dali Kilani retweeted
Sam is honestly 5 steps ahead of everyone. The only reason OpenAI is able to function with all the demand is because of Azure's compute. Sam knows it, and why they signed the deal. But if you read the details, the deal with MSFT specifically only applies to pre-AGI technology So what happens when you achieve AGI? You become a competitor to EVERY SINGLE COMPANY in the world. We will live in a post scarcity world, assuming OAI achieves their goal. You can kill Google, MSFT, etc. However you need 1 thing: hardware. Thats why Sam has been on a fundraising mission in the middle east, specifically recently. Sam knows that you cannot achieve his vision by being dependent on any compute provider. No tech company will provide gpusm, just for you to kill their company You MUST have your own compute and hardware to get an AGI which can help everyone in the world, there is no other way. He's is so far ahead of everyone else lol
124
264
50
2,783
1,600,727
Dali Kilani retweeted
Where did our friends go?? The bike rides. The sleepovers. Your first cigarette. Gossiping about your latest crush. Being consoled when they break your heart. Discovering a new city. Re-discovering your own (at 2AM). The hungover brunches. The tears. The laughs. The love. So many of the moments that shaped us were shared with friends. And then those moments started happening online. Posting photos from last night’s party. Tagging your latest situationship. Poking your roommate (whatever that means??). It was exciting and well intentioned, but something got lost along the way. The platforms that were built to bring friends closer together became places we go to scroll through videos… of strangers. We’re entertained, but where did our friends go? We started amo because we believe that friendship still deserves a place online. We’re a group of friends who have spent our lives building stuff for friends. And now, we’re getting ready to share something new for you and your crew. We call it amo. And to us, amo is more than a company. It’s a friend affair, « une affaire d’amitié ». Something we’re making for you, but for us too. Our first idea is almost here. We hope you love it as much as we love making it. Welcome to amo.
75
100
72
887
523,845
Today's @langchain launch is HUGE!!!!!! Try if for yourself chat.langchain.com/ So many are building a way to search & ask questions of company documents and knowledge as a "service desk" interaction. It's not only an employee productivity boost, but also can give your users a better experience while saving hundreds of support hours. We give you the starter code here: github.com/langchain-ai/chat…
5
11
3
76
20,773
Dali Kilani retweeted
"The people who succeed are irrationally passionate about something." @naval
18
336
11
2,061
103,777
Some @PostgreSQL footguns: - default configuration - long transactions mixed with OLTP workload - repmgr (and other HA tools not based on consensus algorithms) - LISTEN/NOTIFY - "pg_dump is a utility for backing up a PostgreSQL database" says the official doc - moving/copying PGDATA and ignoring glibc version change - hot_standby_feedback on/off dilemma - partitioning of tables having high number of indexes and receiving QPS >> 1k - "setting statement_timeout in postgresql.conf is not recommended because it would affect all sessions" says the official doc - using replication slots w/o setting max_slot_wal_keep_size - relying only on statement_timeout & idle_in_transaction_session_timeout and thinking it's enough (lack of transaction_timeout) - data types "money", "enum", "timestamp" (3 different cases) - int4 PK - massive DELETE - attempts to use transactional version of DDL (e.g., CREATE INDEX) under heavy loads - subtransactions - DDL under load without lock_timeout
23
64
8
529
133,246
🚀 Exploring Llama-2’s Quality: Can we replace generalist GPT-4 endpoints with specialized OSS models? Dive deep with our technical blogpost to understand the nuances and insights of fine-tuning OSS models. 🔗anyscale.com/blog/fine-tunin… 🧵 Thread 1/N👇
15
110
23
518
374,430
6 months ago it looked like AI / LLMs were going to bring a much needed revival to the venture startup ecosystem after a few tough years. With companies like Jasper starting to slow down, it’s looking like this may not be the case. Right now there are 2 clear winners, a handful of losers, and a small group of moonshots that seem promising. Let’s start with the losers. Companies like Jasper and the VCs that back them are the biggest losers right now. Jasper raised >$100M at a 10-figure valuation for what is essentially a generic, thin wrapper around OpenAI. Their UX and brand are good, but not great, and competition from companies building differentiated products specifically for high-value niches are making it very hard to grow with such a generic product. I’m not sure how this pans out but VC’s will likely lose their money. The other category of losers are the VC-backed teams building at the application layer that raised $250K-25M in Dec - March on the back of the chatbot craze with the expectation that they would be able to sell to later-stage and enterprise companies. These startups typically have products that are more focused than something very generic like Jasper, but still don't have a real technology moat; the products are easy to copy. Executives at enterprise companies are excited about AI, and have been vocal about this from the beginning. This led a lot of founders and VC's to believe these companies would make good first customers. What the startups building for these companies failed to realize is just how aligned and savvy executives and the engineers they manage would be at quickly getting AI into production using open-source tools. An engineering leader would rather spin up their own @langchain and @trychroma infrastructure for free and build tech themselves than buy something from a new, unproven startup (and maybe pick up a promotion along the way). In short, large companies are opting to write their own AI success stories rather than being a part of the growth metrics a new AI startup needs to raise their next round. (This is part of an ongoing shift in the way technology is adopted; I'll discuss this in a post next week.) This brings us to our first group of winners — established companies and market incumbents. Most of them had little trouble adding AI into their products or hacking together some sort of "chat-your-docs" application internally for employee use. This came as a surprise to me. Most of these companies seemed to be asleep at the wheel for years. They somehow woke up and have been able to successfully navigate the LLM craze with ample dexterity. There are two causes for this: 1. Getting AI right is a life or death proposition for many of these companies and their executives; failure here would mean a slow death over the next several years. They can't risk putting their future in the hands of a new startup that could fail and would rather lead projects internally to make absolutely sure things go as intended. 2. There is a certain amount of kick-ass wafting through halls of the C-Suite right now. Ambitious projects are being green-lit and supported in ways they weren't a few years ago. I think we owe this in part to @elonmusk reminding us of what is possible when a small group of smart people are highly motivated to get things done. Reduce red-tape, increase personal responsibility, and watch the magic happen. Our second group of winners live on the opposite side of this spectrum; indie devs and solopreneurs. These small, often one-man outfits do not raise outside capital or build big teams. They're advantage is their small size and ability to move very quickly with low overhead. They build niche products for niche markets, which they often dominate. The goal is build a saas product (or multiple) that generates ~$10k/mo in relatively passive income. This is sometimes called "mirco-saas." These are the @levelsio's and @dannypostma's of the world. They are part software devs, part content marketers, and full-time modern internet businessmen. They answer to no one except the markets and their own intuition. This is the biggest group of winners right now. Unconstrained by the need for a $1B+ exit or the goal of $100MM ARR, they build and launch products in rapid-fire fashion, iterating until PMF and cashflow, and moving on to the next. They ruthlessly shutdown products that are not performing. LLMs and text-to-image models a la Stable Diffusion have been a boon for these entrepreneurs, and I personally know of dozens of successful (keeping in mind their definition of successful) apps that were started less than 6 months ago. The lifestyle and freedom these endeavors afford to those that perform well is also quite enticing. I think we will continue to see the number of successful micro-saas AI apps grow in the next 12 months. This could possibly become one of the biggest cohorts creating real value with this technology. The last group I want to talk about are the AI Moonshots — companies that are fundamentally re-imagining an entire industry from the ground up. Generally, these companies are VC-backed and building products that have the potential to redefine how a small group of highly-skilled humans interact with and are assisted by technology. It's too early to tell if they'll be successful or not; early prototypes have been compelling. This is certainly the most exciting segment to watch. A few companies I would put in this group are: 1. cursor.so - an AI-first code editor that could very well change how software is written. 2. harvey.ai - AI for legal practices 3. runwayml.com - an AI-powered video editor This is an incomplete list, but overall I think the Moonshot category needs to grow massively if we're going to see the AI-powered future we've all been hoping for. If you're a founder in the $250K-25M raised category and are having a hard time finding PMF for your chatbot or LLMOps company, it may be time to consider pivoting to something more ambitious. Lets recap: 1. VC-backed companies are having a hard time. The more money a company raised, the more pain they're feeling. 2. Incumbents and market leaders are quickly become adept at deploying cutting-edge AI using internal teams and open-source, off-the-shelf technology, cutting out what seemed to be good opportunities for VC-backed startups. 3. Indie devs are building small, cash-flowing businesses by quickly shipping niche AI-powered products in niche markets. 4. A small number of promising Moonshot companies with unproven technology hold the most potential for VC-sized returns. It's still early. This landscape will continue to change as new foundational models are released and toolchains improve. I'm sure you can find counter examples to everything I've written about here. Put them in the comments for others to see. And just to be upfront about this, I fall squarely into the "raised $250K-25M without PMF" category. If you're a founder in the same boat, I'd love to talk. My DMs are open. If you enjoyed this post, don't forget to follow me, Sam Hogan. I share one long-form post per week covering AI, startups, open-source, and more. That's all folks! Thanks for reading. See you next week.
128
402
161
2,424
1,513,796
Dali Kilani retweeted
The National Cybersecurity Implementation Plan is out! Read the entire thing, it's great, well-written, and relatively short. whitehouse.gov/briefing-room… Because this is Twitter and no one likes to read things, here are some of the highlights:
1
8
1
30
5,225
Introducing LongLLaMA 🦙, an unlimited-context version of OpenLLaMA fine-tuned at 8k & capable of extrapolating to 256k tokens! We train it using our new Focused Transformer 🎯 technique (FoT). No degradation on short context, drop-in compatibility & Apache 2.0 license 🔥🔥 🧵
32
182
23
800
209,535
If you need a swiss army knife Docker image run: docker run --rm -it ghcr.io/teaxyz/infuser It'll auto-install packages when you run commands. 🔥 Magic
1
10
55
9,027