Home
>
Topics
>
OpenAI

Top 25 OpenAI articles on Substack

Latest OpenAI Articles


Brace for Impact: Here Comes the "Cram Down"

Upcoming Edtech Happy Hour Events, ASU+GSV 2024 Session Overviews, US Newspapers Sue OpenAI, Coursera and Chegg Stock Down, and more!
Brace for Impact: Here Comes the “Cram Down” By Ben Kornell
Sarah Morin, Ben Kornell, and Alex Sarlin ∙ 3 LIKES
Matt Rubins
Ben - this is so insightful and so true. Twain said "history doesn't repeat itself, but it often rhymes". I've lived through three of these cycles now - the S&L crisis in '90-93, the dot com and telecom winter from '01 to '04, and then Global Financial Crisis from '08-'11. Every time we go through the same cycle. When a bubble bursts, during the first year people believe that a recovery is right around the corner. It'll be fine! The second year, they realize this may take a while longer and that they need to start cutting costs to extend the runway and avoid exposing themselves to "market pricing discovery". When they run out of moves, they reach the capitulation stage and that's when the dreaded "inside down round" happens. People start to read the deal docs and understand how weighted average anti-dilution provisions really work, what discounts on notes and SAFEs really do to founder economics, and how pay to play provisions work. It's ugly. The companies that get through this phase quickly, or even better proactively in the first two years, are well positioned to be acquirors of both market share and weaker competitors. These cycles typically last 4 years and we're about 18-24 months into this one.
I'm very optimistic about the future. We're seeing strong revenue growth in our portfolio and the long term trends underlying the digitation of education and alternative ways to upskill the workforce are very much intact. It just takes time, but anyone who's been around education for a long time knows that everything takes time in our business.

Google, Perplexity and OpenAI seek to mould Changing Consumer Behaviors

🔎 The Future of Search will be decided in the next five years.
Image: the geeks at Perplexity. Is the internet willing to change their repetitive behaviors? Hello Everyone, As OpenAI prepares to launch a search product that will utilize Bing and GPT-5 to directly compete with Google, something significant is about to happen
Michael Spencer ∙ 41 LIKES

The Sam Altman Playbook

Fear, The Denial of Uncertainties, and Hype
How do you convince the world that your ideas and business might ultimately be worth $7 trillion dollars? Partly by getting some great results, partly by speculating about unlimited potential, and partly by downplaying and ignoring inconvenient truths.
Gary Marcus ∙ 132 LIKES
George Burch
There is a major flaw less said. GenAI can't certify itself. It is a black box and even if it seems to reach a 2 sigma the human labor costs to prove that which are huge cannot guarantee it will respond correctly on the very next prompt. And if it takes more labor to check it than it saves it hardly is a general intelligence.
Raul I Lopez
“all of this has happened before. all of this will happen again.”
Yep, I’ve been there. Working on AI research in 1990-1991, just before the second AI Winter.

Tom White
"What if this was more common? Great writing is precious enough that I wish we had multiple interpretations of most great works. It would be a great way to see the evolution of artists."
Yes! Mark Twain on Jane Austen is a good (read as: hilarious) place to start: In his extensive correspondence with fellow author and critic William Dean Howells, Mark Twain seemed to enjoy venting his literary spleen on Jane Austen precisely because he knew her to be Howells’ favorite author, In 1909 Twain wrote that “Jane Austin” [sic] was “entirely impossible” and that he could not read her prose even if paid a salary to do so. Howells notes in My Mark Twain (1910) that in fiction Twain “had certain distinct loathings; there were certain authors whose names he seemed not so much to pronounce as to spew out of his mouth...
Rather than pitying Twain when he was sick, Howells threatened to come and read Pride and Prejudice to him.
Twain marveled that Austen had been allowed to die a natural death rather than face execution for her literary crimes. “Her books madden me so that I can’t conceal my frenzy,” Twain observed, apparently viewing an Austen novel as a book which “once you put it down you simply can’t pick it up.” ... In a letter to Joseph Twichell in 1898, Twain fumed, “I have to stop every time I begin. Everytime I read “Pride and Prejudice” I want to dig her up and beat her over the skull with her own shin-bone.” From: https://www.vqronline.org/essay/barkeeper-entering-kingdom-heaven-did-mark-twain-really-hate-jane-austen


Last Week in AI #269: Better evals for multimodal AI, new OpenAI lawsuits, Meta's AI ads tool troubles, AI startups focus on enterprise, and more!

Reka AI releases Vibe-Eval, 8 US newspapers sue OpenAI, Meta's AI ads tool's overspending problem, AI startups are pivoting to enterprise customers
Top News Vibe-Eval: A new open and hard evaluation suite for measuring progress of multimodal language models Reka AI introduces Vibe-Eval, a new evaluation suite designed to measure the progress of multimodal language models. Researchers from the company have created a set of challenging prompts to test the capabilities of these models, particularly focu…
Last Week in AI ∙ 3 LIKES

Last Week in AI #268: Gen AI for gene editing, Moderna partners with OpenAI, model releases from Microsoft and Snowflake, and more!

Gen AI used to generate new gene editors like CRISPR, Moderna's internal ChatGPTs, Microsoft releases Phi-3-mini LLM that can run on a phone, Snowflake open sources enterprise LLM
Top News Generative A.I. Arrives in the Gene Editing World of CRISPR Generative AI, which have already revolutionized areas such as art and programming, are now making significant strides in biotechnology. A new A.I. system developed by the Berkeley-based startup Profluent has been designed to create blueprints for novel gene editors by employing methods …
Last Week in AI ∙ 8 LIKES

LWiAI Podcast #165 - Sora challenger, Astribot's S1, Med-Gemini, Refusal in LLMs

China unveils Sora challenger able to produce videos from text similar to OpenAI tool, Capabilities of Gemini Models in Medicine, and more!
Our 165th episode with a summary and discussion of last week's big AI news! Email us your questions and feedback at contact@lastweekin.ai and/or hello@gladstone.ai Subscribe Apple Podcasts Spotify YouTube RSS Timestamps + links: Tools & Apps (00:01:27)
Last Week in AI ∙ 5 LIKES
Stash of Code
There's a fierce and interesting review of Github Copilot Workspace there:

☁️ Amazon: Wild Margin Expansion

AI requires billions in Capex but it looks like money well spent
Welcome to the Friday edition of How They Make Money. Over 100,000 subscribers turn to us for business and investment insights. In case you missed it: 🚖 Tesla: Robotaxi Pivot ♾ Meta: The Anti-Apple 🔎 Google: "A Positive Moment" 🍿 Netflix: Engagement Machine
App Economy Insights ∙ 52 LIKES

Microsoft and OpenAI’s increasingly complicated relationship

An AI Soap Opera in the making?
You might think that Microsoft owns 49% of OpenAI. But as far as I understand it doesn’t. It has a right to about 49% of a for-profit subsidiary of OpenAI’s profits, up until a very complex point that may require litigation to resolve, but the for-profit hasn’t made any profits, and the for-profit is owned by a non-profit. And I’ll be damned if I can ac…
Gary Marcus ∙ 53 LIKES
Gerben Wierda
Is "it's complicated" a civilised way to say 'clusterfuck'? Or might all of thus mean that OpenAI has handed Microsoft the means to fill whatever mini-'moat' OpenAI had? Did OpenAI give away whatever crown jewels they had in that Microsoft deal that got them the compute they needed? Definitely intriguing.
Ko
Relation"shop" haha is that deliberate?

Met Gala Not Dead, But Decaying

The Gala itself has a new theme, a new carpet design, and varied guests each year, but has become rather predictable.
Before the Met Gala Monday night, Anna Wintour apologized for “confusion” over the theme. The Costume Institute exhibition at the Metropolitan Museum of Art that the Gala opens and raises money for is Sleeping Beauties: Reawakening Fashion, but the dress code was “The Garden of Time.” Anna said on
Amy Odell ∙ 109 LIKES
Anne Hjortshøj
It all seemed so joyless.
J.W.
I wonder if part of the reason is that that the event has outgrown itself in a specific kind of way. When you compare it to the Oscars, for instance, there is a point to it all (the actual awards) that the public is part of, because they can view the entire event. The Met Gala just feels so disconnected now because you know the vast majority of attendees don't care about museums (some of them probably don't even know the difference between the Met and MoMA, I'm guessing) or even fully understand why they are there. So it all just feels kind of fake, and the public isn't allowed inside for the party, so it winds up feeling very hollow, despite the absurd star power on the red carpet.
When it was more of a society event, there was a perception of authenticity, because those attending seemed to have a true interest in the institution of the Met. It felt like more of a NYC-specific type of thing that was more closely connected to the museum. I am not sure if any of this made sense, it's kind of word salad, but tldr: the gala is too big, too corporate, and too phony-feeling now to be relevant.

How Perplexity builds product

Johnny Ho, co-founder and head of product, explains how he organizes his teams like slime mold, uses AI to build their AI company, and much more
👋 Hey, Lenny here! Welcome to this month’s ✨ free edition ✨ of Lenny’s Newsletter. Each week I tackle reader questions about building product, driving growth, and accelerating your career. If you’re not a subscriber, here’s what you missed this month:
Lenny Rachitsky ∙ 160 LIKES
Harshal Patil
Love perplexity, use it every day, and Glad to read more about the behind-the-scenes.
Mostafa Fotouhi
I like these articles, I got things that helped me in my career, thanks a lot.
I didn't know Perplexity but I would like to test it

GC, a16z Capture 44% VC fundraising💰, Massive Acquisitions in Software Startups 🛒, Network Effects🕸️

Welcome to The VC Corner, your weekly dose of Venture Capital and Startups to keep you up and running! 🚀 You can now become a premium subscriber and read the full guest posts I share on The VC Corner. Next Saturday, I will have Peter Walker, head of insights at Carta, publishing in my newsletter a deep dive into actual VC Valuations
Ruben Dominguez Ibar ∙ 13 LIKES
Money for Entrepreneurs
So valuable, as always!

How RLHF works, part 2: A thin line between useful and lobotomized

Many, many signs of life for preference fine-tuning beyond spoofing chat evaluation tools.
See part 1 of this series for a textual overview of the 3 stages of RLHF: instruction-tuning, reward modeling, and RL. 17 months on from the release of ChatGPT and we still do not have any fully open-source replications of its fine-tuning process. We’re much further from it than most people think
Nathan Lambert ∙ 21 LIKES

Nobody Likes a Know-It-All: Smaller LLMs are Gaining Momentum

Phi-3 and OpenELM, two major small model releases this week.
Next Week in The Sequence: Edge 391: Our series about autonomous agents continues with the fascinating topic of function calling. We explore UCBerkeley’s research on LLMCompiler for function calling and we review the PhiData framework for building agents.
Jesus Rodriguez ∙ 25 LIKES

Key journalism funder considers becoming invitation-only

Twelve active calls and survival techniques for starving publishers in a post-grant, post-Meta, post-truth media landscape.
Welcome! This week on the Media Finance Monitor: Conversations about media funding, publishing technology, newsroom leadership and more Civitates is considering becoming invitation-only Survival techniques for starving publishers in a post-grant, post-Meta, post-truth media landscape
Peter Erdelyi and Ioana Epure ∙ 2 LIKES

🔮 Can the West wean off from China?; European startups; AI war rooms; fragile societies ++ #472

Hi, I’m Azeem Azhar. In this week’s edition, we explore China’s dominance of the battery supply chain. And in the rest of today’s issue: Need to know: GenAI as a GPT Is generative AI a general-purpose technology? We’ve long believed it to be one, and mounting evidence over the past year contributes to this position.
Azeem Azhar and Nathan Warren ∙ 22 LIKES

What I Read This Week...

Investors are pulling money out of risk assets, Micron receives a multi-billion dollar grant from the U.S. government, and a new House bill makes a TikTok ban more likely
Watch All-In E175 Read our deep dives into Climate, Artificial Intelligence, Healthcare and Space Caught My Eye… Money is being pulled out of equities and junk bonds at the fastest rate in more than a year. Investors are becoming more conservative in their allocations, citing elevated geopolitical risk and potential upside risk to commodity prices and infl…
Chamath Palihapitiya ∙ 67 LIKES
sayandcode
Honestly the part about ppl talking mongoDB sounds compelling to *move to* the bay area. Beats overhearing about mundane shit like how the dog ate the petunias

This is one of the major things wrong with our country

Big story yesterday: a rich guy, who engaged in a years-long scheme of fraud and international law breaking involving laundering money for drug running, child prostitution, terrorism, and Russian oligarchs, got off with a slap on the wrist. This rich guy is a crypto billionaire, Changpeng “CZ” Zhao, the founder and former CEO of the Binance crypto exc…
Lucian K. Truscott IV ∙ 222 LIKES
Margo Howard
There are two things I have no understanding of: anything Bitcoin, and Trump supporters.
(I sort of understand graft, so thanks for that.)
Patris
Fortunes buy privilege. One of which is immunity from justice. Did, does, will do.


AGI is what you want it to be

Certain definitions of AGI are backing people into a pseudo-religious corner.
Artificial general intelligence (AGI) doesn’t need to involve the idea of agency. The term’s three words only indicate a general level of capability, but given its lack of grounding in any regulatory agency or academic community, no one can control what others think it means. The biggest problem with the AGI debate is different folks have different end …
Nathan Lambert ∙ 24 LIKES
Dylan Patel
The real AGI is the friends we made along the way
Oleksii
Of all definitions, I like The Modern Turing Test (Suleyman) the least. Intuitively, it seems to be the most prone to "reward hacking". A model can just discover something which is trivial but super hard for humans (like flash trading).
All other reasonable definitions do seem to require embodiment. But let's consider this thought experiment - imagine human brain completely separated from its body but still alive and able to communicate with us. That is still AGI because brain implements human intelligence, but will we be able to tell? Intuitively, there seems to be connection with Gödel's first incompleteness theorem, e.g will NGI (us) be able to identify AGI?


AI #62: Too Soon to Tell

What is the mysterious impressive new ‘gpt2-chatbot’ from the Arena? Is it GPT-4.5? A refinement of GPT-4? A variation on GPT-2 somehow? A new architecture? Q-star? Someone else’s model? Could be anything. It is so weird that this is how someone chose to present that model.
Zvi Mowshowitz ∙ 20 LIKES
Dr. Y
>“Every college student should learn to train a GPT-2… not the most important thing but I bet in 2 years that’s something every Harvard freshman will have to do”
This used to be called "writing in a diary" back when people did their own thinking.
rational_hippy
Hey Zvi! Thanks a lot for mentioning the Pause AI Protests! I am actually using a Partiful Invite for organising the Paris protest rather than the facebook page: https://partiful.com/e/3Tl1xrS6i9NUZxyJGf5G

🔮 What is Tesla?; LLM gene editor; China rivals GPT-4; crypto, Kant & conscious molluscs ++ #471

Hi, I’m Azeem Azhar. In this week’s edition, we explore all the layers that make up Tesla’s identity. And in the rest of today’s issue: Need to know: Research breakthrough Biotech company Profluent has achieved the first successful precision editing of the human genome using a programmable gene editor designed with large language models.
Azeem Azhar and Nathan Warren ∙ 38 LIKES
steven monahan
Your concise comment hit home. "Linear thinking no longer applies in a world that is changing at an exponential rate. Since 2015, I have been researching, interpreting and discussing how emerging technologies will dramatically change every aspect of our lives.
Your insight is spot on. I had a left brain stroke while producing one of my TEDx Talks event. It immediately took my two left brains off line impacting my linear, logical, sequential thinking. My right non-Linear as I see them"Nous" brains have offset my damaged two left brains, and are wide open online and generating an infinity of ideas faster than I can apply them.
praxis22
Cheers for the Deep Mind doc, shall read

What happened in Marketing: TikTok is Back, IG Algorithm shifts & LinkedIn is 🤐

IAB AdTech launches, Organic on IG & LinkedIn scales, Google AI & Amazon, the ad giant.
Anyone keeping up with Kendrick Lamar vs Drake? Many Brands are Drake, starting with solutions that people want. After Success, they try to not focus on product, instead do the PR and Event runs. You know the Aftermath. Before we begin, You can access the newsletter archive and support my work and Discord community by simply taking an action below:
Jaskaran ∙ 7 LIKES
Martin O'Leary
What are your thoughts on LinkedIn In-app professional games?
Matilda Lucy
Professional games on LinkedIn 🤢