This new DeepMind research shows just how broken vector search is.

Turns out some docs in your index are theoretically incapable of being retrieved by vector search, given a certain dimension count of the embedding.

Plain old BM25 from 1994 outperforms it on recall.

1/4 This result gives me a lot of joy as a search nerd for more than a decade.

Haters will say that the dataset the authors created, LIMIT, is synthetic and unrealistic, but this has been my observation building search systems at Google / Glean.

Source:


2/4alphaxiv.org/pdf/2508.21038

I'm using GPT5 Pro to find me the best stocks and startup investments.

Asked it to use modern portfolio theory and size investments.
—Top Privates [+9.7%]: Databricks, Stripe, Anthropic, SpaceX
—Top Publics [+14.2%]: Nvidia, TSMC, Microsoft, Meta

Just put $1000 into the stocks! Prompt: "Check all public / private stock market companies and tell me what I should invest in from first principles reasoning. You have $1000.

Please do deep research and present rationale for each investment. Each one should have a target price and expected value. Use advanced math for trading. Draw research from authoritative sources like research and unbiased pundits. Size my bets properly and use everything you know about portfolio theory. Corroborate each decision with a list of predictions about those companies.

Your goal is to maximize expected value. Make minimum 5 investments. Write it in a table."

Ridiculous that OpenAI claimed 74.9% on SWE-Bench just to prove they were above Opus 4.1’s 74.5%…

By running it on 477 problems instead of the full 500.

Their system card only says 74% too. Source: cdn.openai.com/pdf/8124a3ce-a… cdn.openai.com/pdf/8124a3ce-a…

🚨Anthropic is at $4.5B annualized revenue and is the fastest growing software company in history!

They just overtook OpenAI to become the market leader in LLM API cost.

We just dropped this and more in our mid year Enteprise AI report:

1/7 Enterprise LLM API spend has exploded from $3.5B to $8.4B by mid year, and that number is already stale!

2/7

Microsoft just leaked their official compensation bands for engineers.

We often forget that you can be a stable high-performing engineer with
great work-life balance, be a BigTech lifer and comfortably retire with a net worth of ~$15M! Source:
archive.is/biMgu

The best open-source AI model just dropped a detailed report on how it was trained, a rare resource for students given no frontier lab is publishing!

Kimi K2's estimated total cost of training is ~$20-30M, roughly in line with pricing: $0.6/M in $2.5/M out tokens.

10 highlights: 1. Generating tokens by rewriting high-quality tokens with LLMs in pre-training
2.  Mining 3000+ MCPs and using LLM-generated personas to improve agentic tool calling
3.  10,000 parallel Kubernetes sandboxes to solve Github issues
4.  New scaling laws for sparsity in MoE models

The hardest high school math exam in the world, the 6 problem 9 hour IMO 2025, was this week.

AI models performed poorly.

Gemini 2.5 Pro scored the highest, just 13/42, costing $431.97, in a best of 32 eval. Bronze cutoff was 19.

Long way to go for AI to solve hard Math. Math Eval: matharena.ai/imo/
Problems: sugaku.net/content/imo-20…

Most important tech blog this year: OpenAI engineer and ex-founder of $3.5B Segment wrote a tell all post about how OpenAI works internally.

From obsession with X, devout use of Slack to engineering culture and tech stack.

A peek under the hood of a generational company. Source: calv.info/openai-reflect…

I'm using the best AI models to bet $1000 on Polymarket!

Asked it to use modern portfolio theory + bet sizing to make calculated bets. It chose everything from BTC price to Fed rates.

Expected returns:
o3-pro: +21.6%
opus 4: +41.7%
grok 4 heavy: +34%

Will report back who won. Wanted to use Gemini 2.5 Pro too but on AI Studio, it did not search the web. I’ve kicked off a Deep Research and will report back under this thread.

Holy shit it’s happening.

OpenAI wants to own ALL the distribution. Source: reuters.com/business/media…

Why are 6' tall well-educated high earning men AND attractive women single?

Modern dating is in a crisis for everyone. It is one of the biggest problems of our time. Let’s break it down with data.

1. Women are more educated than ever, and prefer men the same or more.
Women are getting educated at higher rates than men (47% of women 25-34 have a bachelors vs 37% of men), yet strongly prefer partners of the same education level (81% of couples where one is college-educated have both college-educated). Few men meet this criteria.

2. Women make more money than ever, and prefer men who make more.
College-educated women in cities make MORE than men (women age 25-34 earn >100% of men in 22 large US metros including NYC and DC), yet data shows they are hypergamous, meaning they'd rather date men who make more than them (only 16% of heterosexual marriages have a female breadwinner). Data also shows that women save less than men of commensurate income, meaning the man's income becomes important to keep up their lifestyle if they ever want to purchase a house or settle down. Few men meet this criteria.

3. Women's standards are heightened by dating apps, which force filtering on criteria that are superficial and not long-term focused.
Women are in short supply in dating apps (74% on Tinder, 64% of Hinge is male), and confuse the hookup market for the marriage market. They apply selective filtering given the affordances of the apps (height, looks, education, age, race) and match with men who are "out of their league" and don't want to settle with them (median users message people 10-17%ile more attractive than themselves). When they do want to settle, this group of men self-select for being incompatible on another critical dimension like values, income or personality, leaving women unhappy with the result. Good men who fall short of the initial filter come across too strong and/or feel like a backup and find it hard to love a woman who he can sense is "settling" for him.

4. Social media has polarized both genders.
Both sides have false expectations around dating. For women, they're taught to never settle for less, be an independent girlboss, and that they don't need anyone short of perfection. Men are taught that the only women worth liking having perfect Instagram bodies, are told to date women with fewer sexual partners, that women want them for money, and the world today is just against them in every way. This makes many of them isolate and not enjoy women at all. They see less eye to eye moreso than ever on politics, opinions, and shared interests as a result. The gender “war” in Korea best exemplifies this.

5. Women take longer to find a long term match, but are seen as less attractive with age.
Because of 3, women often date around longer in their most appealing and fertile years. They see a rapid fall off in their own matches after 30 (data shows women desirability peaks at 18-22 while men climbs until 50), when it becomes a race to the bottom of the declining quality of suitors that are interested and their own declining standards: "I should've just been with that one guy from 5yrs ago." When they find a guy, they want to move quickly into marriage and kids which is often too aggressive for men. Although its supposedly taboo to say but men also prefer women with less former partners especially when choosing a spouse, which makes it more difficult for these women (data shows men willingness to commit drops sharply >8 sexual partners)

6. The world is global, and paths diverge more often than before.
Men and women both are increasingly ambitious and will move cities in search of better opportunities. Long distance relationships have lower survival rates, which consume a lot of prime dating years (40% of long distances fail, with an avg separation of 4.5mos).

7. Dating advice reinforces poor behavior.
All of this is exacerbated by people on both sides who are either in a relationship or single offering unhelpful advice like "just suck it up" or "try harder" which makes men and women cement their pre-existing habits. 20%+ of people <30 say their dating guidance primarily comes from TikTok / Instagram.

8. Ambitious women don't want kids more than men. The ones that do struggle to find partners that can afford it.
64% of women under 50 without children don't want kids vs 50% for men. Men often want kids in the former case or can't really afford them in the latter (median cost in metros is $25-32k/yr for a child and 36% of adults cite cost as the major reason for no kids). Either way, it's a bad situation.

Neither women nor men are at fault for this situation society has found itself in, and this post isn’t blaming anyone in particular. The cause is an intricate combination of the state of the economy, access through dating apps, social media and innate mating preferences. If we cannot talk Sources:
bls.gov/charts/usual-w…
robinhood.org/news/the-true-…
pewresearch.org/short-reads/20…
pewresearch.org/social-trends/…
mix999fm.iheart.com/content/2024-0…
luvlink.com/blogs/news/sur…
pubmed.ncbi.nlm.nih.gov/27805420/
bigthink.com/culture-religi…
theblog.okcupid.com/a-womans-advan…
martinvidal.medium.com/dating-apps-ar…

This 67 page AI report on how 300 execs at software startups like Cursor, ElevenLabs, Sierra (revenue $10m-$1b+) use AI just dropped.

I read it all so you don't have to. Top 7 takeaways:

OpenAI is still the #1 model provider in the enterprise, but Claude is second.

1/10 On AI spend.

Companies are spending more on data storage, processing and AI infra than inference and training (quite surprising to me)! Of course, AI talent is by far the most expensive line item.

2/10

Venture Capital Compensation in the US

Many people kept telling me they don't know how VC comp works, so here it is, split by fund size, based on 2024 survey data, 500+ samples. The reason the base (white) + bonus (yellow) doesn't add up to total (green) is because they're all medians, rounded.

We just dropped our consumer AI report today, surveying 5000+ US adults on how they use AI in their daily lives.

I read all 25 pages so you don’t have to. Here are the top 10 learnings:

Nearly 30% of people dating and half of personal coding projects use AI!

1/10 1.8B users use AI every month but only 3% pay today, spending $12B on it.

2/10

No one talks about the real reason driving the ~500k tech layoffs.

Section 174 in the 2017 tax cuts turned engineer salaries from an instant tax deduction into a 5yr write off, causing billions in tax bills.

It even incentivizes offshoring R&D, which has a 15yr write off!

1/4 The numbers are staggering:
— Microsoft: Paid extra $4.8 BILLION in taxes, fired 16,000+ workers
— Meta: Cut 21,000 employees (25% of workforce) after "material" tax increases
— Amazon: 27,000 layoffs concentrated in R&D teams
— Google: 12,000 cuts despite record profits

2/4

Karpathy today said Cursor for Slides needs to exist.. but it already does.

When asked to "Create a detailed data-driven slide deck based on Google's recent financial filings",

It created a stunning 6 page deck with graphs, diagrams IN Google theme!

It's called Genspark. It thinks about the plan of each slide, and it does this unique thing where it generates graphics in Python (matplotlib) and then moves the assets into a landscape static HTML website that it compiles into a slide.

Pretty neat tactic.

The BEST AI report in the world just dropped and I read all 340 pages so you don't have to.

Must-read for frankly, everybody. Here are the top 20 learnings from the report:

ChatGPT has 8x'd growth in 2.5yrs to 800M monthly users.
India at 14% is the #1 largest user base.

1/20 ChatGPT hit 1B searches a day 5.5x quicker than Google.

2/20

If these 3 Japanese companies you've never heard of went down, all modern digital infrastructure (and every AI breakthrough) would grind to a halt.

Here's the story of JSR, Tokyo Ohka Kogyo (TOK) and Shin-Etsu, the most important companies you've never heard of...

1/7 These 3 control 80% of the global EUV photoresist market.

Photoresists are exotic light-sensitive chemicals that enable the patterning of circuits on silicon wafers when UV or EUV light hits these materials through a mask.

How did they achieve this dominance?

2/7

🚨 This was the BEST Google I/O that I can remember.

Google launched over 12 different insane things.

Here is every single one of the launches and the best tweets about them:

1/12 The future of building software.

LLMs are pretty good at generating code, but they're slow. Gemini Diffusion is 10-15x faster than autoregressive models by using diffusion, which used to be for images. This is the 2nd model after Mercury Small to show this.

2/12

https://twitter.com/1297627095330242561/status/1924908812009668915

🚨 BigTech jobs have ZERO growth in the last 3yrs.

Google, Microsoft, Apple, Tesla, Meta, Nvidia and Palantir - the biggest tech employers have collectively stagnated headcount, data shows.

This is why CS majors can’t get jobs. Bigtech hypergrowth era is over.

Here's why:

1/4 — They overhired in the pandemic (2020-2022) during zero interest rates
— Revenue growth of several consumer software products peaked in the pandemic but slowed after it was over
— Higher interest rates force CFOs to juice operating margin

2/4

🚨Viral rumors of DeepSeek R2 leaked!

—1.2T param, 78B active, hybrid MoE
—97.3% cheaper than GPT 4o ($0.07/M in, $0.27/M out)
—5.2PB training data. 89.7% on C-Eval2.0
—Better vision. 92.4% on COCO
—82% utilization in Huawei Ascend 910B

Big shift away from US supply chain. Source:
jiuyangongshe.com/h5/article/1h4…