04-19 08:02 · AI,技术,HackerNews,AI
Show HN: AI Subroutines – Run automation scripts inside your browser tab
Most web agents solve the wrong half of the problem. You can get an LLM to post on X, DM on Instagram, or send a LinkedIn connection request — once. The moment you need to do it a thousand times, the economics break: tokens per invocation, latency per invocation, non-determinism per invocation. On outreach, CRM updates, and bulk posting, "the agent clicked the wrong button this time" is not a quirk. It's a failure mode.
The obvious fix is to skip the UI and call the site's internal API directly. That's correct, and it's where most "just call the API" projects die. Because the hard problem isn't the endpoint. It's auth.
Authenticated web requests carry some combination of cookies, rotating CSRF tokens, session tokens, bearer headers, anti-replay nonces, fingerprint-bound parameters, and request-signing hashes computed in the site's own JS at request time. Some are set by the server. Some are derived in the browser. Some rotate per request.
Out-of-process scrapers — Node workers, Playwright workers, cloud functions — have to rebuild all of that out of band. That's the thing that breaks the moment a site rotates a header or ships a new signing scheme. Most HAR-replay tooling ends its useful life right here.
In rtrvr, both the recording and the replay happen inside the user's browser, from within the webpage itself.
No Puppeteer driver. No headless worker. No separate TLS stack. The browser does what it always does: attach the cookies, run the site's own JS to compute the headers, ship the request.
Auth, CSRF, signing, and fingerprinting all propagate for free.The agent never touches any of it. No key extraction, no session rebuild, no proxy rotation.
This sounds like a footnote. It's the whole architecture.
There's a second problem hiding inside "just record the network." A typical minute of browsing fires dozens to hundreds of requests per tab — analytics beacons, RUM pings, feature-flag polls, third-party pixels, prefetches, media chunks, hot-module reload pokes. TheactualAPI call you care about is often 3 requests out of 300.
You cannot hand all of that to an LLM to figure out which one is the tool. It does not fit in the context window, and even if you paid to stretch it, the signal drowns in the noise.
So before the generator sees anything, we rank and trim the capture. Requests are scored on a handful of weighted signals:
Concretely: a first-party mutatingPOSTthat fires 80ms after a click with a 200 response and a body lands around +83. A generic analytics beacon is −80. Everything in between gets ordered and the top five survive. Those five plus the DOM interactions around them get rendered into a 12 000-character context for the generator; if it overruns, we drop visited URLs first, then network candidates, then DOM hints, and re-render until it fits.
Even after ranking, a strong candidate is not automatically replay-worthy. If the top request carries a volatile operation identifier — X'squeryId, Meta'sdoc_id, any GraphQL operation hash pinned to the current deploy — the planner forces a DOM-only tool regardless of score, and the generator is instructed not to surface those values in the first place (Do NOT expose or discover queryId/doc_id/operationHash values). This is the single most useful failure case to catch early: network replay looks great in a demo and breaks quietly a week later when the site ships.The docsgo deeper on how the DOM / network / hybrid decision is made and thertrvr.*helper namespace the generated code uses.
This is the unglamorous step that makes recording→Subroutine actually work. In-page execution solves auth for free; ranked trimming — with the volatile-ID circuit breaker — is what lets the generator reliably pick the right request to templatize.
A recorded task — aSubroutine— is registered as a callable tool in the agent's tool set, next tosearchandfetch:
Point the agent at a sheet of 500 rows. It picks parameters per row. The Subroutine runs. The LLM is invoked exactly once per row — for parameter selection — and the action itself is a script.
A Subroutine is a small async JavaScript function that runs in the tab. The parameters the agent passes — the row from the sheet, the target URL, the message body — are injected asconstdeclarations above your code. Inside the body, anrtrvr.*helper namespace covers the common moves you need on real sites without dropping down to brittle selectors or hand-rolledfetchscaffolding:
A minimal LinkedIn "connect" Subroutine:
DOM when the UI is the stable contract,rtrvr.requestwhen the endpoint is. The generator mixes them as needed, and because everything runs in the page, cookies and CSRF tokens are just there — you read them, you don't rebuild them.
A few non-obvious implementation details worth knowing:
We record HTTP. Sites that do real work over WebSockets, WebRTC, or heavy mid-flow client-side derivation need DOM actions interleaved into the Subroutine. That path is slower and less reliable — we treat it as a fallba
▸ 展开全文
04-19 08:00 · AI,技术,HackerNews,AI
Dad brains: How fatherhood rewires the male mind
Dad brains: How fatherhood rewires the male mind
From before their babies are born, men undergo serious hormonal changes that can powerfully influence their behaviour – with consequences for their child's wellbeing.
In the months before my son was born, my partner and I attended an active birth workshop, a breastfeeding session and the hospital-run antenatal course, read a small pile of pregnancy and baby books and scrolled through loads of websites. Our notepads quickly filled up.
Among my notes of that time are details of the many ways women's bodies prepare for birth and motherhood: hormones rise and drop, organs move, brains reshape.
No one, however, told me that my brain and body were also readying for fatherhood.
My son was over a year old when I first came across that idea in Father Time, a book by primatologist Sarah Blaffer Hrdy in which she argues that men have all the necessary biological wiring to be "every bit as protective and nurturing as the most committed mother".
This piqued my curiosity. I am a resolute believer in active fathering, but I had imagined this was a cultural decision by my generation of men. Hrdy's book, however, introduced me to an entire academic field saying that our approach is rooted in biology, just dormant and waiting to be triggered.
After interviewing Hrdy and other experts and delving into the studies, I came to a simple conclusion: fatherhood changes men in ways that echo how motherhood transforms women. The more involved a father is with their baby's care, the deeper this transition becomes. These shifts in our endocrine and neural system show that the nurturing father is not a modern aberration, but a deeply rooted biological trait.
Falling testosterone
The earliest research on how fathers are physically changed by babies came from observations of other animals. These late 20th-Century studies found that many mammalian males – including other primates – show clear hormonal shifts, including rises and drops in hormones like testosterone, vasoprin and prolactin, typically associated with motherhood, as they engage in active parental care.
When the American anthropologist Lee Gettler, then an undergrad student, heard about these findings in the early 2000s, he was hooked.
"I asked [my lecturer] whether anyone was studying these questions in human fathers, and the answer at that point was largely no", says Gettler, now the director of the Hormones, Health, and Human Behavior Laboratory at the University of Notre Dame in Indiana.
The first-ever study demonstrating hormonal changes in men had just been published in 2000 by two Canadian scholars – Katherine Wynne-Edwards and Anne Storey. By the time Gettler looked into this field, it was already an established fact that fathers had lower testosterone that men without kids.
"But there's a chicken and the egg problem there, right?" Gettler explained to me. "Are low testosterone men more likely to become fathers? Or does the transition to fatherhood kind of lead to this cascade of biological changes in men?"
To answer this question and others, Gettler teamed up with the scientists running a decades-long project in Cebu City, Philippines.
In 2005, this team collected saliva samples from 624 men, with an average age of 21 years old and without partners, and tested them for testosterone, then four years later tested them again. They wanted to answer two questions: would men that become fathers in the interim have lower testosterone, and would it be even lower in fathers that spent more hours doing childcare?
When the results came back, the answer to both questions was "yes". The men that had babies showed significantly lower levels of testosterone compared to non-fathers. And the men that had spent longer looking after babies showed the largest drops in testosterone. Those that shared a bed with their infants also had lower levels.
"I think it was the first clear message in the scientific literature that men have this capacity to prepare for fatherhood," Gettler told me. In a way, he explains, this is their biology preparing them for caregiving.
Their findings are not unique. Other teams have also found that drops in testosterone during their partner's pregnancy are also linked with higher investment, commitment and satisfaction after birth, and that this hormone’s level was even linked to the men's reactions to baby cries: it made them more alert and responsive. In 2018, a team in Gettler's lab also concluded that fathers with lower levels of testosterone tend to be more involved in caring for babies and toddlers.
But when does this happen? The question of whether it's before or after birth was bubbling in the mind of James K Rilling, the director of the Laboratory for Human Social Neuroscience at Emory University in the US.
"My assumption," Rilling told me, "was that it would happen during the postnatal period after fathers spent some time interacting with their infants."
What they found surprised them. When they tested exp
▸ 展开全文
04-19 07:59 · AI,技术,HackerNews,AI
College instructor turns to typewriters to curb AI-written work
The scene is right out of the 1950s with students pecking away at manual typewriters, the machines dinging at the end of each line.
Once each semester, Grit Matthias Phelps, a German language instructor at Cornell University, introduces her students to the raw feeling of typing without online assistance. No screens, online dictionaries, spellcheckers or delete keys.
The exercise started in spring 2023 as Phelps grew frustrated with the reality that students were using generative AI and online translation platforms to churn out grammatically perfect assignments.
“What’s the point of me reading it if it’s already correct anyway, and you didn’t write it yourself? Could you produce it without your computer?” said Phelps.
She wanted students to understand what writing, thinking and classrooms were like before everything turned digital. So, she found a few dozen old manual typewriters in thrift shops and online marketplaces, and created what her syllabus calls an “analog” assignment.
It might be premature to say that typewriters are making a comeback beyond Cornell’s campus. But the revival is part of a national trend toward old-school testing methods like in-class pen-and-paper exams and oral tests to prevent AI use for assignments on laptops.
Typewriters bring ‘old days’ taste of doing one thing at a time
Students arrived for class on a recent analog day to find typewriters at the desks, some with German and some with QWERTY keyboards.
“I was so confused. I had no idea what was happening. I’d seen typewriters in movies, but they don’t tell you how a typewriter works,” said Catherine Mong, 19, a freshman in Phelps’ Intro to German class. “I didn’t know there was a whole science to using a typewriter.”
Like a rotary phone, the manual typewriter appears simple but is not intuitive to the smartphone generation. Phelps demonstrated how to feed the paper manually, striking the keys with force but not so hard the letters would smudge. She explained that the dinging bell signifies the end of a line and the need to manually return the carriage to start the next line. (“Oh,” said one student, “that’s why it’s called ‘return.'”)
“Everything slows down. It’s like back in the old days when you really did one thing at a time. And there was joy in doing it,” said Phelps, who brings in her two children, aged 7 and 9, to serve as “tech support” and ensure no one has their phones out.
Students welcomed having fewer distractions
The assignment carries lessons beyond simply how to use a typewriter, which is the whole point.
“It dawned on me that the difference with typing on a typewriter is not just how you interact with the typewriter, but how you interact with the world around you,” said computer science major Ratchaphon Lertdamrongwong, a sophomore, whose class had to write a critique of a German movie they’d watched.
In the absence of screens, there are no notifications to distract you as you write. Without every answer readily available at his fingertips, he asked his classmates for help, which Phelps heartily encouraged.
“While writing the essay, I had to talk a lot more, socialize a lot more, which I guess was normal back then,” Lertdamrongwong said, referring to the typewriter era. “But it’s drastically different from how we interact within the classroom in modern times. People are always on a laptop, always on the phone.”
Without a delete key and the ability to correct every mistake, he paused to think more intentionally about his writing.
“This might sound bad, but I was forced to actually think about the problem on my own instead of delegating to AI or Google search,” he said.
Manual machines were a workout for pinky fingers
Most students found their pinkies weren’t strong enough to touch-type, so they typed more slowly, pecking at the keyboard with their index fingers.
Mong, the freshman, faced the added challenge of a recently broken wrist, requiring her to use just one hand. The self-described perfectionist was initially frustrated with how messy her page looked with odd spacing between certain letters and misspellings. (Phelps told students to backspace and type ‘X’s over errors.)
“This thing I handed in had pencil marks all over it and definitely did not look clean or finished. But it’s part of the process of learning that you’re going to make mistakes,” said Mong, who found the assignment of typing a poem “fun and challenging.”
She embraced the odd spacing and played with the visual boundaries of the page to indent and fragment lines in the style of poet E.E. Cummings. It took several sheets of paper and many mistakes, all of which Mong saved.
“I’m probably going to hang them on my wall,” Mong said. “I’m kind of fascinated by typewriters. I told all my friends, I did a German test on a typewriter!”
The Associated Press’ education coverage receives financial support from multiple private foundations. AP is solely responsible for all content. Find AP’s standards for working with philanthropie
▸ 展开全文
04-18 08:02 · AI成本,技术栈风险,运营成本,商业模式,规模化挑战
Are the costs of AI agents also rising exponentially? (2025)
There is an extremely important question about the near-future of AI that almost no-one is asking.
We’ve all seen the graphs from METR showing that the length of tasks AI agents can perform has been growing exponentially over the last 7 years. While GPT-2 could only do software engineering tasks that would take someone a few seconds, the latest models can (50% of the time) do tasks that would take a human a few hours.
As this trend shows no signs of stopping, people have naturally taken to extrapolating it out, to forecast when we might expect AI to be able to do tasks that take an engineer a full work-day; or week; or year.
But we are missing a key piece of information — the cost of performing this work.
Over those 7 years AI systems have grown exponentially. The size of the models (parameter count) has grown by 4,000x and the number of times they are run in each task (tokens generated) has grown by about 100,000x. AI researchers have also found massive efficiencies, but it is eminently plausible that the cost for the peak performance measured by METR has been growing — and growing exponentially.
This might not be so bad. For example, if the best AI agents are able to complete tasks that are 3x longer each year and the costs to do so are also increasing by 3x each year, then the cost to have an AI agent perform tasks would remain the same multiple of what it costs a human to do those tasks. Or if the costs have a longer doubling time than the time-horizons, then the AI-systems would be getting cheaper compared with humans.
But what if the costs are growing more quickly than the time horizons? In that case, these cutting-edge AI systems would be getting less cost-competitive with humans over time. If so, the METR time-horizon trend could be misleading. It would be showing how the state of the art is improving, but part of this progress would be due to more and more lavish expenditure on compute so it would be diverging from what is economical. It would be becoming more like the Formula 1 of AI performance — showing what is possible, but not what is practical.
So in my view, a key question we need to ask is:
How is the ‘hourly’ cost of AI agents changing over time?
By ‘hourly’ cost I mean the financial cost of using an LLM to complete a task right at the model’s 50% time horizon divided by the length of that time horizon. So as with the METR time horizons themselves, the durations are measured not by how long it takes the model, but how long it typically takes humans to do that task. For example, Claude 4.1 Opus’s 50% time horizon is 2 hours: it can succeed in 50% of tasks that take human software engineers 2 hours. So we can look at how much it costs for it to perform such a task and divide by 2, to find its hourly rate for this work.
I’ve found that very few people are asking this question. And when I ask people what they think is happening to these costs over time, their opinions vary wildly. Some assume the total cost of a task is staying the same, even as the task length increases exponentially. That would imply an exponentially declining hourly rate. Others assume the total cost is also growing exponentially — after all, we’ve seen dramatic increases in the costs to access cutting-edge models. And most people (myself included) had little idea of how much it currently costs for AI agents to do an hour’s software engineering work. Are we talking cents? Dollars? Hundreds of dollars? An AI agent can’t cost more per hour than a human to complete these tasks can it? Can it?
⁂
A couple of months ago I asked METR if they could share the cost data for their benchmarking. I figured it would be easy — just take the cost of running their benchmark for each model, plot it against release date and see how it is growing. Or plot the cost of each model vs its time horizon and see the relationship.
But they helpfully pointed out that it isn’t so easy at all. Their headline time-horizon numbers are meant to show the best possible performance that can be attained with a model (regardless of cost). So they run their models inside an agent scaffold until the performance has plateaued. Since they really want to make sure it has plateaued, they use a lot of compute on this and don’t worry too much about whether they’ve used too much. After all, if you are just trying to find the eventual height of a plateau, there is no problem in going far into the flat part of the graph.
But if you are trying to find out when the plateau begins, there is a problem with this strategy. Their total spend for each model is sometimes just enough to get onto the plateau and sometimes many times more than is needed. So total spend can’t be used as direct estimate of the costs of achieving that performance.
Fortunately, they released a chart that can be used to shed some light on the key question of how hourly costs of LLM agents are changing over time:
This chart (from METR’s page for GPT-5) shows how performance increases with cost. The cost in
▸ 展开全文
04-18 08:00 · SEO优化,内容发布,技术教程,开发者社区,网站架构
How to Host a Blog on a Subdirectory Instead of a Subdomain
In this guide, you’ll learn how to host your blog on a subdirectory (e.g. example.com/blog) instead of a subdomain (e.g., blog.example.com). Every step here has been tested and verified to work.
Introduction
Hosting your blog on a subdirectory can improve SEO and enhance user experience.
Although there are a lot of articles that espouse the benefits of using subdirectories over subdomains, few resources that provide a step-by-step guide on how to actually set this up.
Why Host on a Subdirectory?
The benefits to hosting on a subdirectory is primarily to improve SEO.
There are a lot of other articles out there on this topic, but they all say something similar to the following:
-
Hosting your blog on a subdirectory is better for SEO because it consolidates your website’s authority and ranking power.
-
Google has stated that they do not treat subdomains as a separate entity.
-
Despite what Google has stated, empiric data suggests that subdirectories outperform subdomains in search rankings.
-
If you want to maximize your SEO efforts, hosting on a subdirectory is the way to go.
If you want to learn more, you can read this article by ButterCMS: Blog Subdomain or Subdirectory? Hint: One is 40% Better.
My personal experience has been similar. When I moved a blog from a subdomain to a subdirectory, I saw a noticeable increase in organic traffic and search engine rankings. The increase happened after a few weeks. During that time, I did not release any new content and nor did I promote the blog.
Why Not Host on a Subdirectory?
The setup is more complex. Many blogging platforms and CMSs are designed to work on subdomains, and configuring them to work on a subdirectory can be tricky.
I’ve personally found the setup process to be quite time consuming. It’s a tricky process and you have to follow the instructions carefully. After having previously changed a blog from a subdomain to a subdirectory, I’ve found difficult to justify the time to do it for taikohub.com.
If you still think it’s worth your time, then read on.
Steps to Host Your Blog on a Subdirectory
Lets suppose you have two sites right now. One is example.com and the other is blog.example.com. You want to host the blog on example.com/blog instead of blog.example.com.
Lets also suppose your blog (blog.example.com) is a Next.js app hosted on Vercel and your main site (example.com) is a static site hosted on Render.
Although Vercel and Render are used as examples here, the steps are nearly identical for other hosting providers. You do not need configure anything for your hosting provider. Everything can be done from the Cloudflare Dashboard, and from the comforts of your text editor.
Important: Note that Cloudflare often changes their dashboard UI and routes. If you find that the steps here do not match what you see on your Cloudflare Dashboard, just use the search function in the dashboard to find the relevant section.
Step 1: Set Up DNS Records for the Main Site
First, set up the DNS records for your main site (example.com). Again, if you do not use Render, then follow the equivalent steps for your hosting provider. Generally this should be in their documentation.
-
Go to your Cloudflare Dashboard. Click into your domain, then SSL/TLS, then Overview. Then click “Configure”.
-
Next, select “Custom SSL/TLS” then select “Full”.
-
Go to DNS records in the sidebar by clicking on “DNS”, then “DNS Records”. Then click “Add Record”.
-
Add the following DNS Records. Replace
my-site.onrender.com
with the service URL for your main site. If you have other applications such as an API, you can add those as well. Note that it’s important you set the “Proxy status” to “Proxied”. It’s also important you do NOT add a wildcard record (eg.*.example.com
).
Step 2: Set Up DNS Records for the Blog
-
Make sure your blog is already accessible on a subdomain (eg. blog.example.com).
-
Add another DNS Record for the blog. Replace
cname.vercel-dns.com
with the CNAME target provided by the hosting provider for your blog.
Step 3. Configure Your Next.js Blog
Ensure Correct Routing for Static Assets
-
Make sure that your Next.js blog’s router points to
/
and not/blog
. You should NOT have any routes that contain/blog
. Edit thenext.config.js
ornext.config.mjs
file and addbasePath: "/blog"
to the config./** @type {import('next').NextConfig} */ const nextConfig = { basePath: "/blog", // Add this line images: { remotePatterns: [ { protocol: "https", hostname: "imagedelivery.net", }, ], }, redirects: async () => { return []; }, }; export default nextConfig;
Step 4. Add a Cloudflare Worker
-
Go to Cloudflare Dashboard. Click “Workers & Pages”. Click “Create” then click “Create Worker”.
-
For the purpose of this blog post, we’ll go with the easiest option by selecting “Start with Hello World!”. For production applications, consider using Git. It looks like the following. Lets name it
blog-worker
. Then click “Deploy”.// worker.js /** * Welcome to Cloudflare Workers! This is your first worker. * * -
▸ 展开全文
04-17 10:49 · AI,技术,HackerNews,AI
The beginning of scarcity in AI
For the first time since the 2000s, technology companies are confronting the limits of their supply chain.
GPU rental prices for Nvidia’s Blackwell chips hit $4.08 per hour this week, up 48% from $2.75 just two months ago.1CoreWeave raised prices 20% & extended minimum contracts from one year to three.1
“We’re making some very tough trades at the moment on things we’re not pursuing because we don’t have enough compute.” - Sarah Friar, OpenAI CFO1
This scarcity is already reshaping access. Anthropic has limited its newest model to roughly forty organizations.2Access to the bleeding edge is becoming a gated privilege, for both capacity & security.
If the largest AI companies are having problems, startups face a tougher proposition. Five hallmarks define this era :
The age of abundant AI is over, & it will remain so for years.3
Wall Street Journal, “AI Is Using So Much Energy That Computing Firepower Is Running Out,” April 2026.↩︎↩︎↩︎
tomtunguz.com/mythos↩︎
tomtunguz.com/what-if-we-run-out-of-capacity↩︎
▸ 展开全文
04-17 10:49 · AI,技术,HackerNews,AI
Cloudflare's AI Platform: an inference layer designed for agents
AI models are changing quickly: the best model to use for agentic coding today might in three months be a completely different model from a different provider. On top of this, real-world use cases often require calling more than one model. Your customer support agent might use a fast, cheap model to classify a user's message; a large, reasoning model to plan its actions; and a lightweight model to execute individual tasks.
This means you need access to all the models, without tying yourself financially and operationally to a single provider. You also need the right systems in place to monitor costs across providers, ensure reliability when one of them has an outage, and manage latency no matter where your users are.
These challenges are present whenever you’re building with AI, but they get even more pressing when you’re building agents. A simple chatbot might make one inference call per user prompt. An agent might chain ten calls together to complete a single task and suddenly, a single slow provider doesn't add 50ms, it adds 500ms. One failed request isn't a retry, but suddenly a cascade of downstream failures.
Since launching AI Gateway and Workers AI, we’ve seen incredible adoption from developers building AI-powered applications on Cloudflare and we’ve been shipping fast to keep up! In just the past few months, we've refreshed the dashboard, added zero-setup default gateways, automatic retries on upstream failures, and more granular logging controls. Today, we’re making Cloudflare into a unified inference layer: one API to access any AI model from any provider, built to be fast and reliable.
One catalog, one unified endpoint
Starting today, you can call third-party models using the same AI.run() binding you already use for Workers AI. If you’re using Workers, switching from a Cloudflare-hosted model to one from OpenAI, Anthropic, or any other provider is a one-line change.
const response = await env.AI.run('anthropic/claude-opus-4-6',{
input: 'What is Cloudflare?',
}, {
gateway: { id: "default" },
});
For those who don’t use Workers, we’ll be releasing REST API support in the coming weeks, so you can access the full model catalog from any environment.
We’re also excited to share that you'll now have access to 70+ models across 12+ providers — all through one API, one line of code to switch between them, and one set of credits to pay for them. And we’re quickly expanding this as we go.
You can browse through our model catalog to find the best model for your use case, from open-source models hosted on Cloudflare Workers AI to proprietary models from the major model providers. We’re excited to be expanding access to models from Alibaba Cloud, AssemblyAI, Bytedance, Google, InWorld, MiniMax, OpenAI, Pixverse, Recraft, Runway, and Vidu — who will provide their models through AI Gateway. Notably, we’re expanding our model offerings to include image, video, and speech models so that you can build multimodal applications
Accessing all your models through one API also means you can manage all your AI spend in one place. Most companies today are calling an average of 3.5 models across multiple providers, which means no one provider is able to give you a holistic view of your AI usage. With AI Gateway, you’ll get one centralized place to monitor and manage AI spend.
By including custom metadata with your requests, you can get a breakdown of your costs on the attributes that you care about most, like spend by free vs. paid users, by individual customers, or by specific workflows in your app.
const response = await env.AI.run('@cf/moonshotai/kimi-k2.5',
{
prompt: 'What is AI Gateway?'
},
{
metadata: { "teamId": "AI", "userId": 12345 }
}
);
Bring your own model
AI Gateway gives you access to models from all the providers through one API. But sometimes you need to run a model you've fine-tuned on your own data or one optimized for your specific use case. For that, we are working on letting users bring their own model to Workers AI.
The overwhelming majority of our traffic comes from dedicated instances for Enterprise customers who are running custom models on our platform, and we want to bring this to more customers. To do this, we leverage Replicate’s Cog technology to help you containerize machine learning models.
Cog is designed to be quite simple: all you need to do is write down dependencies in a cog.yaml file, and your inference code in a Python file. Cog abstracts away all the hard things about packaging ML models, such as CUDA dependencies, Python versions, weight loading, etc.
Example of a cog.yaml
file:
build:
python_version: "3.13"
python_requirements: requirements.txt
predict: "predict.py:Predictor"
Example of a predict.py
file, which has a function to set up the model and a function that runs when you receive an inference request (a prediction):
from cog import BasePredictor, Path, Input
import torch
class Predictor(BasePredictor):
def setup(self):
"""Load the model into memory to make running multiple p
▸ 展开全文
04-17 10:49 · AI,技术,HackerNews,AI
Substrate AI Is Hiring Harness Engineers
Substrate (www.substrate.cc) is building the world’s first AI native BPO, starting with healthcare RCM. We’re focused on helping healthcare enterprises like physician groups, EMR providers, and BPOs accelerate AR, and get reimbursed quickly and cost effectively. We already touch over 500k healthcare claims each month.
We're looking for an exceptional Harness Engineer to help build the systems around our agents and AI products. You'll be instrumental in advancing AI systems that are already in production, and improving success rates for probabilistic products, and precision for deterministic ones. You’ll build systems that can understand complex healthcare contracts, read EDI companion guides, utilize tools for navigating highly sensitive healthcare infrastructure and banking/financial infrastructure, and make intelligent decisions about claims processing - all in service of helping healthcare providers get paid fairly.
Some general color on what we’re looking for: https://writing.kunle.app/p/come-work-with-me-at-substrate
Meaningful salary and equity available.
Substrate is building browser based AI agents for healthcare RCM. Our goal here is to augment and replace employees, not replace any particular software a provider uses. We’ll go after the $14b outpatient RCM market, and if we’re successful, providers will finally start getting paid fairly.
We’re the right team for this problem with deep healthcare, AI & fullstack engineering expertise, multiple startups under our belt between us, and a deep network in healthcare that will help bootstrap the first customers and real revenue.
▸ 展开全文
04-17 00:02 · 基础设施更新,云服务,网络安全,企业服务,产品发布
Cloudflare Email Service
Email is the most accessible interface in the world. It is ubiquitous. There’s no need for a custom chat application, no custom SDK for each channel. Everyone already has an email address, which means everyone can already interact with your application or agent. And your agent can interact with anyone.
If you are building an application, you already rely on email for signups, notifications, and invoices. Increasingly, it is not just your application logic that needs this channel. Your agents do, too. During our private beta, we talked to developers who are building exactly this: customer support agents, invoice processing pipelines, account verification flows, multi-agent workflows. All built on top of email. The pattern is clear: email is becoming a core interface for agents, and developers need infrastructure purpose-built for it.
Cloudflare Email Service is that piece. With Email Routing, you can receive email to your application or agent. With Email Sending, you can reply to emails or send outbounds to notify your users when your agents are done doing work. And with the rest of the developer platform, you can build a full email client and Agents SDK onEmail hook as native functionality.
Today, as part of Agents Week, Cloudflare Email Service is entering public beta, allowing any application and any agent to send emails. We are also completing the toolkit for building email-native agents:
Email Sending binding, available from your Workers and the Agents SDK
A new Email MCP server
Wrangler CLI email commands
Skills for coding agents
An open-source agentic inbox reference app
Email Sending: now in public beta
Email Sending graduates from private beta to public beta today. You can now send transactional emails directly from Workers with a native Workers binding — no API keys, no secrets management.
export default {
async fetch(request, env, ctx) {
await env.EMAIL.send({
to: "[email protected]",
from: "[email protected]",
subject: "Your order has shipped",
text: "Your order #1234 has shipped and is on its way."
});
return new Response("Email sent");
},
};
Or send from any platform, any language, using the REST API and our TypeScript, Python, and Go SDKs:
curl "https://api.cloudflare.com/client/v4/accounts/{account_id}/email-service/send" \
--header "Authorization: Bearer <API_TOKEN>" \
--header "Content-Type: application/json" \
--data '{
"to": "[email protected]",
"from": "[email protected]",
"subject": "Your order has shipped",
"text": "Your order #1234 has shipped and is on its way."
}'
Sending email that actually reaches inboxes usually means wrestling with SPF, DKIM, and DMARC records. When you add your domain to Email Service, we configure all of it automatically. Your emails are authenticated and delivered, not flagged as spam. And because Email Service is a global service built on Cloudflare's network, your emails are delivered with low latency anywhere in the world.
Combined with Email Routing, which has been free and available for years, you now have complete bidirectional email within a single platform. Receive an email, process it in a Worker, and reply, all without leaving Cloudflare.
For the full deep dive on Email Sending, refer to our Birthday Week announcement. The rest of this post describes what Email Service unlocks for agents.
Agents SDK: your agent is email-native
The Agents SDK for building agents on Cloudflare already has a first-class onEmail hook for receiving and processing inbound email. But until now, your agent could only reply synchronously, or send emails to members of your Cloudflare account.
With Email Sending, that constraint is gone. This is the difference between a chatbot and an agent.
Email agents receive a message, orchestrate work across the platform, and respond asynchronously.
A chatbot responds in the moment or not at all. An agent thinks, acts, and communicates on its own timeline. With Email Sending, your agent can receive a message, spend an hour processing data, check three other systems, and then reply with a complete answer. It can schedule follow-ups. It can escalate when it detects an edge case. It can operate independently. In other words: it can actually do work, not just answer questions.
Here's what a support agent looks like with the full pipeline — receive, persist, and reply:
import { Agent, routeAgentEmail } from "agents";
import { createAddressBasedEmailResolver, type AgentEmail } from "agents/email";
import PostalMime from "postal-mime";
export class SupportAgent extends Agent {
async onEmail(email: AgentEmail) {
const raw = await email.getRaw();
const parsed = await PostalMime.parse(raw);
// Persist in agent state
this.setState({
...this.state,
ticket: { from: email.from, subject: parsed.subject, body: parsed.text, messageId: parsed.messageId },
});
// Kick off long running background agent task
// Or place a message on a Queue to be handled by another Worker
// Reply here or in other Worker handler, like a Queue handler
await this.sendEmail({
bi
▸ 展开全文
04-17 00:02 · AI安全,技术争议,社区舆论,概念质疑,风险认知
AI cybersecurity is not proof of work
The proof of work is the wrong analogy: finding hash collisions, while exponentially harder with N, is guaranteed to find, with enough work, some S so that H(S) satisfies N, so an asymmetry of resources used will see the side with more "work ability" eventually winning. But bugs are different: 1. Different LLMs executions take different branches, but eventually the possible branches based on the code possible states are saturated. 2. If we imagine sampling the model for a bug in a given code M times, with M large, eventually the cap becomes not "M" (because of saturated state of the code AND the LLM sampler meaningful paths), but "I", the model intelligence level. The OpenBSD SACK bug easily shows that: you can run an inferior model for an infinite number of tokens, and it will never realize(*) that the lack of validation of the start window, if put together with the integer overflow, then put together with the fact the branch where the node should never be NULL is entered regardless, will produce the bug. So, cyber security of tomorrow will not be like proof of work in the sense of "more GPU wins"; instead, better models, and faster access to such models, will win. * Don't trust who says that weak models can find the OpenBSD SACK bug. I tried it myself. What happens is that weak models hallucinate (sometimes causally hitting a real problem) that there is a lack of validation of the start of the window (which is in theory harmless because of the start < end validation) and the integer overflow problem without understanding why they, if put together, create an issue. It's just pattern matching of bug classes on code that looks may have a problem, totally lacking the true ability to understand the issue and write an exploit. Test it yourself, GPT 120B OSS is cheap and available. BTW, this is why with this bug, the stronger the model you pick (but not enough to discover the true bug), the less likely it is it will claim there is a bug. Stronger models hallucinate less, so they can't see the problem in any side of the spectrum: the hallucination side of small models, and the real understanding side of Mythos.
▸ 展开全文