Peter Yang
Recently hit 100K subscribers on YouTube.
Some reflections:
1. Packaging (e.g., thumbnails and titles) matters but not as much as delivering value right away.
2. Viewers can tell if your podcast guest is trying to promote something. Sharing practical workflows matters more than how famous the guest is.
3. Some of my videos have 300K+ views while others have 1K. I can't predict the outcome, but I can show up consistently every week.
4. There's a tension between maximizing views ("This model changes everything!") and actually making content that I like. Haven't quite figured this out yet.
Thanks for your support and let me know in the replies what videos you'd like to see next on my channel: https://www.youtube.com/@PeterYangYT
Recently hit 100K subscribers on YouTube.
Some reflections:
1. Packaging (e.g., thumbnails and titles) matters but not as much as delivering value right away.
2. Viewers can tell if your podcast guest is trying to promote something. Sharing practical workflows matters more than how famous the guest is.
3. Some of my videos have 300K+ views while others have 1K. I can't predict the outcome, but I can show up consistently every week.
4. There's a tension between maximizing views ("This model changes everything!") and actually making content that I like. Haven't quite figured this out yet.
Thanks for your support and let me know in the replies what videos you'd like to see next on my channel: https://www.youtube.com/@PeterYangYT
If you want to play my chess engine (WIP): https://qwen-chess.replit.app/
It already seems to perform better than frontier models on chess.
It’s fine-tuned on 2M stockfish-labeled positions, then short GRPO RL pass.
Documentation/tutorial with all the experiments and annotated code.
Amjad Masad: Vibe Research
Fine-tuning a Qwen-8b model to play chess on Replit. Running 3 parallel branches with different experiments and making real progress.
It's amazing how far models have come in their ability to do ML (they used to be really bad at it). So now someone with good
Apparently the distillation model can outperform the teacher model 😂
Arena.ai: Big news: Kimi-K3 by @Kimi_Moonshot is now #1 in the Frontend Code Arena with 1679 pts, surpassing Claude Fable 5.
This is a 17-place jump from Kimi-k2.6 (#18 -> #1).
In Frontend, Kimi-K3 ranked #1 in 6 of 7 domains: Brand & Marketing, Reference-Based Design, Data & Analytics,
Griffin (Griff) Lee 🌉🌁
Supervisor Fielder states, “Drug tolerant housing suggests that the city allows people to openly use drugs in city-funded Permanent Supportive Housing and that’s just not the case.”
Serious? Jackie - Are you ok? 26% of all overdose deaths have occurred in Permanent Supportive Housing since 2020.
Cc: @Gina_McDee and @TheVOSF has the data.
bluoz: SFBOS expanding drug free permanent supportive housing 7.14.26
Supervisor @JackieFielder_ "the term drug tolerant housing suggests the city allows drugs to be openly used, and that's not the case'
followed by supervisor @mattdorsey who reads the 2016 California law that says
It’s super weird to me that air conditioners are not the norm in Europe.
Is there a good explanation why?
Claude getting low-key snarky with me when I questioned the need for a clamp:
"If the branch prediction on an always-true clamp offends you"
Its actual justification was correct: while not mathematically necessary, the paths have different reductions, so the guard is still necessary to avoid a possible sqrt(-epsilon).
I just smiled at the gaslighting as if a kernel invocation and full parameter read-write was going to be a no-op like an always-true scalar C branch. :-)
we make sure some of the top YC AI companies are featured at AIE every single year.
this year, we were graced by @garrytan and @eve_bouff to cap off our startups and design engineer focused audiences respectively. Really enjoyed these raw high value perspectives!
AI Engineer: 🆕 We're so excited to release special double header talks with @ycombinator leaders:
@garrytan on Gbrain, Gstack, and the new Physics of Business: https://www.youtube.com/watch?v=eBUyTS7SzV4
@eve_bouff on Imagination Engineering: https://www.youtube.com/watch?v=Z2Erdirpudo
enjoy!
Ankit Gupta
Congrats to Bunkerhill Health for the latest raise! Tune in to hear @nish_khandwala tell the story of how he went from being a Stanford CS student to building one of the defining companies deploying AI to healthcare.
Y Combinator: Bunkerhill Health (YC S20) has raised $55M to build a state-of-the-art AI platform for health systems. They help hospitals deploy AI across dozens of use cases through a single platform, making it easier to turn new ideas into real patient care.
In this episode of Founder
Sasha Rush
I circled all the Transformer parts of k3. (I'm a sore loser).
Sasha Rush: Time for everyone’s favorite game! Is it a Transformer?
Go to Settings -> Preferences -> Appearance -> Dock icon and Codex lives on.
Ankit Gupta
FYI that America’s moronic visa policy is almost certainly a factor in why Kimi/Moonshot is a Chinese startup and not an American one.
The fact that we don’t staple a green card to every AI PhD completed in America is stupid.
would be more logical to seize their passports and force them to stay
Russ Salakhutdinov: Congratulations to Zhilin Yang, founder and CEO of @Kimi_Moonshot, on the latest Kimi release. What a huge win for the open-source community!
It feels like just yesterday Zhilin was graduating from my lab at CMU, jointly co-advised with William Cohen. Not only did he complete
ajay
~600 PRs merged in two weeks 😅
building loops = the most fun I've had in years
Amjad Masad: Something strange happened at Replit in the past six months.
The same engineers 3x’d output. Support resolved its hardest tickets 60% faster. Anyone could suddenly query the business like an analyst.
We’re seeing a new kind of organization: the self-driving company.
Underrated reply 🤣
Republicans against Trump
Document Declassified by Trump Reveals Russia Tried To Help Him Win the 2020 Election
https://www.mediaite.com/politics/trump/document-declassified-by-trump-reveals-russia-tried-to-help-him-win-the-2020-election/
Please join @garryslist
We are trying to organize the common sense builders
It’s going to be the defining fight for California and America
Mike Solana: for anyone in tech with money: the annual percentage they want to take from you each year via asset seizure is what you should be plowing into super pacs, NGOs, and media influencers right now to beat back marxism. the hour's late, and nobody else is coming to save you.
$NVDA should be pumping on the K3 news. Instead it’s down.
ChatGPT Work is great for tasks in your personal life:
Andrew Ambrosino: the “work” in ChatGPT Work is describing what the agent is doing for you– you can use it for anything you want
I have it reading all of my personal emails, making calendar events, organizing docs for myself and the kids, etc
team is responding to feedback and iterating quickly. we ❤️ our users, thank you all!
Tibo: Evening! We’ve gotten lots of great feedback on the new ChatGPT desktop app (which we didn't get totally quite right on the first try), and as a result, we've made some changes.
1/ ChatGPT conversation history and projects are now visible in the sidebar. Also, your Chat and Work
HOLY FUCKING SHIT the react avengers have assembled
Guillermo Rauch: I’m excited to welcome two legends of developer tools, Pete Hunt (@floydophone) and Nick Schrock (@schrockn), to Vercel.
Pete was one of the pioneers of @reactjs at Meta. He made an early bet to power Instagram Web with ⚛️ React, evangelizing it internally and externally. He
Just be a YC startup
Problem solved
brett goldstein: we need BookFace for non-YC startups
founder friend got ghosted for 6 weeks
after 4 meetings and a bunch of customer diligence convos
VC only responded when a portco founder nudged them, but suggested she meet with an analyst (and answer same questions again)
she said no obvi
Aaron Rupar
to hear Trump tell it, American elections were completely corrupted while he was president and he was helpless to stop it, then they were cleaned up by Biden ahead of Trump's victory in 2024, and now that Trump is president again they are once again becoming third world-caliber
Anonymous
Trump crying about rigged elections, but remember when he was literally recorded trying to rig an election?
How the fuck is this traitor the president of the USA? What the fuck is wrong with you people?
wanye
These people are much more clear-eyed about their project than centrists give them credit for. They aren't interested in the current system, in operating according to its rules or logic. They are in a power struggle for resources, which they want to take from the people with power now by any means necessary (including violence).
Stu Smith: 🚨 Meet the “Abolish Rent” Ideologue Teaching DSA Tenant Organizers How to Wage War on Landlords
Tracy Rosenthal co-founded the Los Angeles Tenants Union, helps politically educate DSA tenant organizers, and is currently on rent strike in New York City.
In this supercut,
Brian Roemmele
Kimi K3 is a gut punch to Anthropic. Its over.
A 2.8T open-weight-bound Chinese model just took the WebDev human-preference crown at launch, posted stronger real agentic terminal and long-horizon coding numbers than Claude Fable 5 on the tasks that actually ship products, and did it at roughly one-third the price while soon to be FREE promising full weights in ten days.
The fear-based marketing that convinced so many to accept closed models, sky-high rents, and “safety” as justification for locking away frontier capability now looks like expensive theater.
Mad are the politicians and academics that fell for Dario hand waving holier than tho testimony that “we must stop AI”. Kimi just drops and the world did not end. The sky did not fall Dario.
Builders no longer have to pretend the only responsible path runs through Anthropic’s premium tollbooth; the open frontier just proved it can match or beat them where it counts and hand the weights to anyone with a cluster.
That narrative is dead.
Brian Roemmele: http://x.com/i/article/2078110904613240832
I'd love to see your best scheduled task to let Codex (or another harness) manage your email and calendar. Please share it in the replies.
Here’s mine. It lists open email action items, unsubscribe candidates, and meeting briefs for me to review.
📌 I talk more about using this in my recent video: https://www.youtube.com/watch?v=WLg9qWOf6zw&t=570s
Peter Yang: ChatGPT Work and Codex are growing by 1M users EVERY FEW DAYS and will likely hit 10M this week.
Here's my new tutorial that walks through how I use Work to do (almost) everything, including how to:
→ Choose the right GPT-5.6 model
→ Organize long-running threads
→ Manage
Peter Yang
I'd love to see your best scheduled task to let Codex (or another harness) manage your email and calendar. Please share it in the replies.
Here’s mine. It lists open email action items, unsubscribe candidates, and meeting briefs for me to review.
📌 I talk more about using this in my recent video: https://www.youtube.com/watch?v=WLg9qWOf6zw&t=570s
Peter Yang: ChatGPT Work and Codex are growing by 1M users EVERY FEW DAYS and will likely hit 10M this week.
Here's my new tutorial that walks through how I use Work to do (almost) everything, including how to:
→ Choose the right GPT-5.6 model
→ Organize long-running threads
→ Manage
Faisal Khan
We were paying close to US$ 270 per month just for our CRM, and with accounting and Invoicing, add another $250 per month. Limited to four users. Build our own, with @Replit and hosted it ourselves, costs us less than $100 per month, and it will eventually save us over $30,000 per year, once we port everything over. Open-Source/Self-Built is the way to go.
If you are over 40 and have the room for it this is genuinely one of the best purchases you can make for mental and physical health.
Francois Chaubard
new pod format!
we simplify the math behind world modeling, rewinding back to optimal control to today's modern AI systems.
Y Combinator: Why do even our best AI models need tens of thousands of examples to learn skills that a human picks up in a handful of tries?
Solving this problem is one of the great open challenges in modern AI. World models, which give AI an internal simulation of its environment, are one of
AI Engineer
Re 🆕 Research to Reality with Google DeepMind
https://youtu.be/1P1hJ36rxM0
Way back in the ancient times of 2018, Google X started a skunkworks project exploring how ML could improve coding. @benoitschilling was CTO, and many people were incredible skeptical.
Benoit now leads the Gemini Thinking, Reasoning, and Coding teams at @GoogleDeepMind, and still sees more potential for AI to solve software - even now that coding has moved from machine-constrained to design-constrained: The challenge is no longer writing the code itself, but ensuring the code aligns with specific, complex requirements, shifting the role of the developer toward architecture, verification, and high-level reasoning.
it seems that the new metagame to boost your x articles is to quote tweet it with an aie video so that your article gets the 500k views from reposting the video on x
i honestly dont hate it at all, just credit the speaker (eg @cormacb) and link the original youtube in the first reply
thanks for tagging us on it
h100envy: Google engineer explained how to fine-tune a tiny LLM from 46% to 90% accuracy on your phone in 21 minutes - better than $1500 on-device AI bootcamps.
pick Gemma 270M -> generate synthetic task data -> fine-tune with LoRA -> quantize to int4 -> deploy to Pixel and hit 2000
Morgan J. Freeman
😆 THE *SHOCK* THAT BORED THE NATION 😆
Trump’s Primetime Meltdown: He Promised a “Huge Shocker” but Delivered Recycled Lies and a Pathetic Failed Voter Suppression Act Push
Desperate Don's National Address was a National Disgrace. It was a total nothingburger. No new evidence, no shocker, just the same recycled 2020 lies that were debunked via 61 lost court cases.
Major networks embarrassed or ghosted him entirely — ABC, NBC, CNN said nah. He’s already threatening to revoke licenses through his corrupt FCC lackey Brendan Carr. His BIG push for his SAVE MY ASS ACT fell flat, and he's exposed to a massive shellacking come the November Midterms.
TRIGGER ALERT FOR MAGA: Proceed with caution.
Full Free Substack Takedown Here ➡http://tinyurl.com/2abv57ah
Blake Scholl 🛫
The current @ycombinator batch is very impressive. Almost half the founders are building hardware.
Back in my batch there were only 3 deep tech companies - @boomsupersonic, @Astranis and @relativityspace.
Ankit Gupta
excited for this new format!
@FrancoisChauba1 and I do a deep dive into world models, mathematical derivations and all.
We’re starting a new series focused on these types of technical topics. Stay tuned!
Y Combinator: Why do even our best AI models need tens of thousands of examples to learn skills that a human picks up in a handful of tries?
Solving this problem is one of the great open challenges in modern AI. World models, which give AI an internal simulation of its environment, are one of
Ned Dwyer
"WATCH USERS AND CRINGE AND THEN FIX IT."
Garry Tan: The point of a startup is to make usable technology for others. When you make software, you have to watch at least 10 people use it. Sit next to them and say absolutely nothing. Force yourself to marinate in the failure of your product design.
Every version 1 of any software
Where did my dear Fable go 😭
Datacurve
Kimi K3 debuts at #3 on DeepSWE.
It's the first open-weights model that delivers frontier-level performance, achieving results similar to Claude Fable and GPT-5.6 Sol.
Aravind Srinivas
At its peak, Sun Microsystems was valued at 205B (394B if inflation adjusted). Sold software in enterprise servers. Got disrupted by Linux, x86, and commodity hardware. Ended up selling to Oracle for 7.4B, losing 96% of its value.
Open source models running on local hardware can have a similar impact given what’s going on.
Marcel Haas
One of the reason which I choose @Replit as my daily AI Dev engine, is the focus on security. Security is integrated deeply in many platform layers. From now on, the Package Firewall is on by default and prevents supply chain attacks. A big topic many were suffering from lately. Replit users who didn't sideload stuff, were not impacted back then and are better protected now.
Replit ⠕: Blocking the bad and removing the noise.
Our @SocketSecurity partnership blocks ~8,000 malicious packages a day and filters out 30% of false-positive critical vulnerability alerts, the ones that were never exploitable.
We focus on security so you can focus on building.
Michele Catasta
We made useful intelligence accessible to every Replit employee, and it fundamentally changed the way we run the company.
Next, we will make useful intelligence accessible to everyone.
Here’s to the next ten years of Replit 🥂
Amjad Masad: http://x.com/i/article/2077626320205574144
brryant
attn: OAI, Anthropic engineers soon to have 8, 9, or 10 figures in your bank account —
You've made it. Congrats. Maybe your immigrant parents will finally congratulate you. But likely not.
Here's what I would do after the shares hits your brokerage account.
GPT-5.6 Sol is the state of the art in cyber. Seeing significant results in applying it to finding and fixing novel vulnerabilities.
Sign up as a defender to use it to secure your systems:
https://openai.com/daybreak/
AI Security Institute: On our cyber range "The Last Ones", GLM-5.2 matches Opus 4.5, released ~7 months before it, while DeepSeek’s V4-Pro falls below Sonnet 4.5, from ~7 months before it.
Sol gets the thing done:
tobi lutke: OpenAI shipped /goal not too long ago.
I feel like GPT-5.6-sol is the first model that just doesn't need it anymore? It just keeps on going until the thing is done. Really impressive.
Laude Institute
Andy @andykonwinski and @swyx talked through why AI researchers may be uniquely well-positioned to wield power responsibly right now, why most of them are still thinking about impact only in terms of companies and citations - and how to change that.
this is cool:
https://welcome-to-codex.openai.chatgpt.site/
btw if you havent set your {codex | claude | gemini | devin} automations to autoresearch how to improve your seo/aeo every week you are really truly missing out on free, should-be-commoditizing-but-weirdly-untapped alpha
SemiAnalysis
Similar to the panic over DeepSeek R1, some uneducated people think Kimi K3’s use of linear attention (KDA) is bad for NVIDIA, HBM, DRAM, and networking because it has relatively lower KV-cache requirements. The opposite is true, and we explain why below. 👇️ 1/8🧵
don't sleep on terra!
Peter Steinberger 🦞: 5.6 Terra high is underrated. Switched @clawsweeper (GitHub review bot) to it and it's ~40% faster overall with negligible quality loss. Better than 5.5 on all counts. Massively cheaper.
(Tried xhigh but that negates perf wins, didn't make a noticable difference in review evals)
RT Rodney Brooks
