Large Language Models Think Too Fast to Explore Effectively
16 by bikenaga | 0 comments on Hacker News.
16 by bikenaga | 0 comments on Hacker News.
Friday, 31 January 2025
Thursday, 30 January 2025
New top story on Hacker News: Show HN: Distr – open-source distribution platform for on-prem deployments
Show HN: Distr – open-source distribution platform for on-prem deployments
14 by louis_w_gk | 0 comments on Hacker News.
Distr is designed to help software engineers distribute and manage their applications or agents in customer-controlled or shared-responsibility environments. You only need a Docker Compose file or Helm chart—everything else for on-prem is handled by the platform. We’re are an open source dev tool company. Over the past couple of months, we’ve spoken with dozens of software companies to understand their challenges with on-prem deployments. We analyzed the internal tools they’ve built and the best practices from existing solutions, combining them into a prebuilt, Open Source solution that works out of the box and integrates seamlessly. Distr consists of two key components: 1. Hub - Provides a centralized view of all deployments and controls connected agents. - Comes with a simple GUI but also supports API and SDK access for seamless integration. - Fully Open- Surce and self-hostable, or you can use our fully managed platform. 2. Lightweight Agents - Pre-built agents for Helm (Kubernetes) and Docker Compose (VM) that run alongside your application. - Handle lifecycle tasks like guided installation, updates, and rollbacks. - Provide basic metrics (health status, application version) and logs If you already have a customer portal or self-service interface for on-prem deployments, you can seamlessly integrate all features into your existing portal or application using our API or SDK. Alternatively, you can use our pre-built, white-labeled customer portal. Here’s what an integration into your existing customer portal could look like: import {DistrService} from "@glasskube/distr-sdk"; const customerHasAutoUpdatesEnabled = false; // replace with your own logic const deploymentTargetId = 'da1d7130-bfa9-49a1-b567-c49728837df7'; const service = new DistrService({ apiKey: 'distr-8c24167aeb5fd4bb48b6d2140927df0f' }); const result = await service.isOutdated(deploymentTargetId); if(result.deploymentTarget.deployment?.latestStatus?.type !== 'ok') { // let the user decide whether to allow updates from an instable state, e.g. with: if(!confirm('The deployment is not in a stable state. Do you want to update anyway?')) { return; } } if(result.outdated) { if(customerHasAutoUpdatesEnabled) { await service.updateDeployment({deploymentTargetId}); // notify customer about the update } else { const newerVersionsAvailable = result.newerVersions; // notify customer about the newer versions, e.g. via email } } With the SDK/API, you can: - Display real-time deployed version and deployment status directly within the application, notifying customers when their deployed version is outdated. - Allow customers to trigger updates from within your app using a simple API call If you’re distributing software and want to streamline updates or enhance monitoring, we’d love your feedback and are here to answer any questions. Getting started is easy—just bring your Docker Compose file or Helm chart, and we’ll guide you through the rest. Check out the fully managed version ( https://ift.tt/E7TiJxb ) and explore our documentation ( https://distr.sh/docs/ ) to learn more.
14 by louis_w_gk | 0 comments on Hacker News.
Distr is designed to help software engineers distribute and manage their applications or agents in customer-controlled or shared-responsibility environments. You only need a Docker Compose file or Helm chart—everything else for on-prem is handled by the platform. We’re are an open source dev tool company. Over the past couple of months, we’ve spoken with dozens of software companies to understand their challenges with on-prem deployments. We analyzed the internal tools they’ve built and the best practices from existing solutions, combining them into a prebuilt, Open Source solution that works out of the box and integrates seamlessly. Distr consists of two key components: 1. Hub - Provides a centralized view of all deployments and controls connected agents. - Comes with a simple GUI but also supports API and SDK access for seamless integration. - Fully Open- Surce and self-hostable, or you can use our fully managed platform. 2. Lightweight Agents - Pre-built agents for Helm (Kubernetes) and Docker Compose (VM) that run alongside your application. - Handle lifecycle tasks like guided installation, updates, and rollbacks. - Provide basic metrics (health status, application version) and logs If you already have a customer portal or self-service interface for on-prem deployments, you can seamlessly integrate all features into your existing portal or application using our API or SDK. Alternatively, you can use our pre-built, white-labeled customer portal. Here’s what an integration into your existing customer portal could look like: import {DistrService} from "@glasskube/distr-sdk"; const customerHasAutoUpdatesEnabled = false; // replace with your own logic const deploymentTargetId = 'da1d7130-bfa9-49a1-b567-c49728837df7'; const service = new DistrService({ apiKey: 'distr-8c24167aeb5fd4bb48b6d2140927df0f' }); const result = await service.isOutdated(deploymentTargetId); if(result.deploymentTarget.deployment?.latestStatus?.type !== 'ok') { // let the user decide whether to allow updates from an instable state, e.g. with: if(!confirm('The deployment is not in a stable state. Do you want to update anyway?')) { return; } } if(result.outdated) { if(customerHasAutoUpdatesEnabled) { await service.updateDeployment({deploymentTargetId}); // notify customer about the update } else { const newerVersionsAvailable = result.newerVersions; // notify customer about the newer versions, e.g. via email } } With the SDK/API, you can: - Display real-time deployed version and deployment status directly within the application, notifying customers when their deployed version is outdated. - Allow customers to trigger updates from within your app using a simple API call If you’re distributing software and want to streamline updates or enhance monitoring, we’d love your feedback and are here to answer any questions. Getting started is easy—just bring your Docker Compose file or Helm chart, and we’ll guide you through the rest. Check out the fully managed version ( https://ift.tt/E7TiJxb ) and explore our documentation ( https://distr.sh/docs/ ) to learn more.
Wednesday, 29 January 2025
New top story on Hacker News: A minimal PyTorch implementation for training your own small LLM from scratch
A minimal PyTorch implementation for training your own small LLM from scratch
17 by amrrs | 1 comments on Hacker News.
17 by amrrs | 1 comments on Hacker News.
New top story on Hacker News: DeepSeek's Hidden Bias: How We Cut It by 76% Without Performance Loss
DeepSeek's Hidden Bias: How We Cut It by 76% Without Performance Loss
44 by nicolevin | 16 comments on Hacker News.
44 by nicolevin | 16 comments on Hacker News.
Tuesday, 28 January 2025
Monday, 27 January 2025
Sunday, 26 January 2025
Saturday, 25 January 2025
Friday, 24 January 2025
Thursday, 23 January 2025
New top story on Hacker News: Show HN: I built an active community of trans people online
Show HN: I built an active community of trans people online
64 by t4t | 25 comments on Hacker News.
A year ago I surveyed the internet and noticed there was only one popular space for trans and gender-non-conforming people to meet; Lex. Lex is not well liked by its users. Its software feels heavy and it is full of cash grabs and anti-patterns. It was recently acquired and is sure to only become more hostile to its users as it turns towards profit generation. With this in mind I built t4t, an alternative specially designed for not only queer people, but specifically trans people. It is an extremely lightweight service. I built it with my most ideal stack: Flutter, Svelte, Supabase, Posthog. It has grown in the last year to about 4,000 monthly active users. I think it could grow way beyond that this year.
64 by t4t | 25 comments on Hacker News.
A year ago I surveyed the internet and noticed there was only one popular space for trans and gender-non-conforming people to meet; Lex. Lex is not well liked by its users. Its software feels heavy and it is full of cash grabs and anti-patterns. It was recently acquired and is sure to only become more hostile to its users as it turns towards profit generation. With this in mind I built t4t, an alternative specially designed for not only queer people, but specifically trans people. It is an extremely lightweight service. I built it with my most ideal stack: Flutter, Svelte, Supabase, Posthog. It has grown in the last year to about 4,000 monthly active users. I think it could grow way beyond that this year.
Wednesday, 22 January 2025
Tuesday, 21 January 2025
Monday, 20 January 2025
Sunday, 19 January 2025
Saturday, 18 January 2025
New top story on Hacker News: Show HN: ZX Spectrum SCR to PNG Converter
Show HN: ZX Spectrum SCR to PNG Converter
3 by iamflimflam1 | 0 comments on Hacker News.
Scratching my own itch. I had to do this for showing information on ZX Spectrum games. So thought I'd turn it into a useful tool for other people to use.
3 by iamflimflam1 | 0 comments on Hacker News.
Scratching my own itch. I had to do this for showing information on ZX Spectrum games. So thought I'd turn it into a useful tool for other people to use.
Friday, 17 January 2025
Thursday, 16 January 2025
Wednesday, 15 January 2025
New top story on Hacker News: Show HN: I Built a Fair Alternative to Product Hunt for Indie Makers Like You
Show HN: I Built a Fair Alternative to Product Hunt for Indie Makers Like You
11 by lakshikag | 3 comments on Hacker News.
I’m an indie maker, just like many of you. A few months back, I launched a product on one of the big platforms, and... nothing. It got buried under dozens of other launches within hours. All that work, all that excitement is gone in the blink of an eye. No one even saw it. It stung. I wasn’t mad, well, maybe a little but mostly, I just felt invisible. The truth is, indie makers like me don’t have big teams or budgets to fight for visibility. We rely on genuine support and connections. I couldn’t stop thinking about how many great ideas never get the attention they deserve because they’re overshadowed. So, I decided to build something different: https://itslaunched.com Here’s the idea: • 10 launches per day, max. Limiting the number of daily launches ensures that every product gets its moment in the spotlight. • 2 votes per user, per day. This isn’t a popularity contest. You only get two votes, so people have to really think about which products they want to support. It’s quality over quantity. • “Under Radar” feature. This one’s my favorite. If a product doesn’t get much love on its launch day, it gets a second chance to shine the next day. Because timing shouldn’t be the only thing standing between you and success. There’s more like badges, comments, streaks but the heart of it is simple: a fair shot for indie makers. I built this because I believe every product deserves to be seen, especially the ones built by solo makers and small teams putting their heart into something they truly care about. And I didn’t build this to compete with Product Hunt. I built it to give indie makers the platform they deserve, one where their creativity truly gets noticed. If this sounds like something you’d want to check out, I’d love your thoughts. I’m still tweaking and improving it every day based on feedback. Let me know what you think and if you’ve got a product you’re proud of, I’d love to see it shine.
11 by lakshikag | 3 comments on Hacker News.
I’m an indie maker, just like many of you. A few months back, I launched a product on one of the big platforms, and... nothing. It got buried under dozens of other launches within hours. All that work, all that excitement is gone in the blink of an eye. No one even saw it. It stung. I wasn’t mad, well, maybe a little but mostly, I just felt invisible. The truth is, indie makers like me don’t have big teams or budgets to fight for visibility. We rely on genuine support and connections. I couldn’t stop thinking about how many great ideas never get the attention they deserve because they’re overshadowed. So, I decided to build something different: https://itslaunched.com Here’s the idea: • 10 launches per day, max. Limiting the number of daily launches ensures that every product gets its moment in the spotlight. • 2 votes per user, per day. This isn’t a popularity contest. You only get two votes, so people have to really think about which products they want to support. It’s quality over quantity. • “Under Radar” feature. This one’s my favorite. If a product doesn’t get much love on its launch day, it gets a second chance to shine the next day. Because timing shouldn’t be the only thing standing between you and success. There’s more like badges, comments, streaks but the heart of it is simple: a fair shot for indie makers. I built this because I believe every product deserves to be seen, especially the ones built by solo makers and small teams putting their heart into something they truly care about. And I didn’t build this to compete with Product Hunt. I built it to give indie makers the platform they deserve, one where their creativity truly gets noticed. If this sounds like something you’d want to check out, I’d love your thoughts. I’m still tweaking and improving it every day based on feedback. Let me know what you think and if you’ve got a product you’re proud of, I’d love to see it shine.
Tuesday, 14 January 2025
Monday, 13 January 2025
Sunday, 12 January 2025
Saturday, 11 January 2025
Friday, 10 January 2025
Thursday, 9 January 2025
New top story on Hacker News: Show HN: TabPFN v2 – A SOTA foundation model for small tabular data
Show HN: TabPFN v2 – A SOTA foundation model for small tabular data
22 by onasta | 2 comments on Hacker News.
I am excited to announce the release of TabPFN v2, a tabular foundation model that delivers state-of-the-art predictions on small datasets in just 2.8 seconds for classification and 4.8 seconds for regression compared to strong baselines tuned for 4 hours. Published in Nature, this model outperforms traditional methods on datasets with up to 10,000 samples and 500 features. The model is available under an open license: a derivative of the Apache 2 license with a single modification, adding an enhanced attribution requirement inspired by the Llama 3 license: https://ift.tt/QoplCUI . You can also try it via API: https://ift.tt/I6YdSLB TabPFN v2 is trained on 130 million synthetic tabular prediction datasets to perform in-context learning and output a predictive distribution for the test data points. Each dataset acts as one meta-datapoint to train the TabPFN weights with SGD. As a foundation model, TabPFN allows for fine-tuning, density estimation and data generation. Compared to TabPFN v1, v2 now natively supports categorical features and missing values. TabPFN v2 performs just as well on datasets with or without these. It also handles outliers and uninformative features naturally, problems that often throw off standard neural nets. TabPFN v2 performs as well with half the data as the next best baseline (CatBoost) with all the data. We also compared TabPFN to the SOTA AutoML system AutoGluon 1.0. Standard TabPFN already outperforms AutoGluon on classification and ties on regression, but ensembling multiple TabPFNs in TabPFN v2 (PHE) is even better. There are some limitations: TabPFN v2 is very fast to train and does not require hyperparameter tuning, but inference is slow. The model is also only designed for datasets up to 10k data points and 500 features. While it may perform well on larger datasets, it hasn't been our focus. We're actively working on removing these limitations and intend to release new versions of TabPFN that can handle larger datasets, have faster inference and perform in additional predictive settings such as time-series and recommender systems. We would love for you to try out TabPFN v2 and give us your feedback!
22 by onasta | 2 comments on Hacker News.
I am excited to announce the release of TabPFN v2, a tabular foundation model that delivers state-of-the-art predictions on small datasets in just 2.8 seconds for classification and 4.8 seconds for regression compared to strong baselines tuned for 4 hours. Published in Nature, this model outperforms traditional methods on datasets with up to 10,000 samples and 500 features. The model is available under an open license: a derivative of the Apache 2 license with a single modification, adding an enhanced attribution requirement inspired by the Llama 3 license: https://ift.tt/QoplCUI . You can also try it via API: https://ift.tt/I6YdSLB TabPFN v2 is trained on 130 million synthetic tabular prediction datasets to perform in-context learning and output a predictive distribution for the test data points. Each dataset acts as one meta-datapoint to train the TabPFN weights with SGD. As a foundation model, TabPFN allows for fine-tuning, density estimation and data generation. Compared to TabPFN v1, v2 now natively supports categorical features and missing values. TabPFN v2 performs just as well on datasets with or without these. It also handles outliers and uninformative features naturally, problems that often throw off standard neural nets. TabPFN v2 performs as well with half the data as the next best baseline (CatBoost) with all the data. We also compared TabPFN to the SOTA AutoML system AutoGluon 1.0. Standard TabPFN already outperforms AutoGluon on classification and ties on regression, but ensembling multiple TabPFNs in TabPFN v2 (PHE) is even better. There are some limitations: TabPFN v2 is very fast to train and does not require hyperparameter tuning, but inference is slow. The model is also only designed for datasets up to 10k data points and 500 features. While it may perform well on larger datasets, it hasn't been our focus. We're actively working on removing these limitations and intend to release new versions of TabPFN that can handle larger datasets, have faster inference and perform in additional predictive settings such as time-series and recommender systems. We would love for you to try out TabPFN v2 and give us your feedback!
New top story on Hacker News: Perpetual Movement: Francis Picabia's 391 Review (1917–1924)
Perpetual Movement: Francis Picabia's 391 Review (1917–1924)
4 by prismatic | 0 comments on Hacker News.
4 by prismatic | 0 comments on Hacker News.
Wednesday, 8 January 2025
Tuesday, 7 January 2025
Monday, 6 January 2025
Sunday, 5 January 2025
Saturday, 4 January 2025
New top story on Hacker News: Show HN: Open Rewind – POC for audio and screen and video streaming to S3
Show HN: Open Rewind – POC for audio and screen and video streaming to S3
11 by wwoessi | 1 comments on Hacker News.
Got into a rabbit hole today. POC works using 'npx efficient-recorder'. Is this useful to anyone?
11 by wwoessi | 1 comments on Hacker News.
Got into a rabbit hole today. POC works using 'npx efficient-recorder'. Is this useful to anyone?