Then Microsoft had the episode where some of their Surface hardware would not reliably stay in sleep mode and cooked itself while being transported in a bag. At the time, Microsoft tried to excuse this by claiming that "a fundamental Computer Science problem" needed to be solved to fix this issue. Strange how other manufacturers could do this without overcoming unsolvable problems in frontier CS research.
While I'm usually a die-hard Microsoft fanboi, I have concluded that their Surface line is terrible."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Microsoft builds MacBook Pro rival with NVIDIA-powered Surface Laptop Ultra"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.windowslatest.com/2026/06/01/microsoft-builds-its-ultimate-macbook-pro-rival-with-the-nvidia-powered-surface-laptop-ultra/"}},"_tags":["comment","author_chihuahua","story_48355720"],"author":"chihuahua","children":[48363458,48363507,48364853,48366782,48367068,48368558],"comment_text":"When I worked at Microsoft (years ago), some employees had Surface laptops. They frequently had issues where the laptop just wasn't working right and required rebooting, at the start of a meeting where they wanted to connect the Surface laptop to a projector. Always the Surfaces, never the Lenovos. One of the Surface things split into two parts, the screen (containing the actual computer) and the keyboard. There was something weird about connecting and disconnecting those parts, some motorized docking/undocking mechanism, that caused problems.
Then Microsoft had the episode where some of their Surface hardware would not reliably stay in sleep mode and cooked itself while being transported in a bag. At the time, Microsoft tried to excuse this by claiming that "a fundamental Computer Science problem" needed to be solved to fix this issue. Strange how other manufacturers could do this without overcoming unsolvable problems in frontier CS research.
While I'm usually a die-hard Microsoft fanboi, I have concluded that their Surface line is terrible.","created_at":"2026-06-01T21:57:57Z","created_at_i":1780351077,"objectID":"48363204","parent_id":48363108,"story_id":48355720,"story_title":"Microsoft builds MacBook Pro rival with NVIDIA-powered Surface Laptop Ultra","story_url":"https://www.windowslatest.com/2026/06/01/microsoft-builds-its-ultimate-macbook-pro-rival-with-the-nvidia-powered-surface-laptop-ultra/","updated_at":"2026-06-15T01:40:12Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"lihanc111"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["frontiercs"],"value":"Evolving agent systems (agents that use continual learning, memory, or context refinement) are becoming the mainstream. However, up until now, most papers show gains on only about 10 tasks, or small subsets of existing benchmarks.
The bigger issue is benchmark saturation. We highlight "circle packing" in the post as a prime example: four different recent papers (ThetaEvolve, TTT-Discover, AdaEvolve, and EvoX) all converge to the exact same ceiling (2.635983). Once multiple methods collapse to the same endpoint because they are essentially just wrapping the same SLSQP Python script, the benchmark stops measuring open-ended improvement.
To fix the evaluation gap, we built Frontier-CS. It consists of 172 algorithmic tasks and 68 research tasks, designed by ICPC World Finalists and CS PhDs.
Our main goal was to create environments with massive search spaces that require actual information-efficient exploration and long-horizon reasoning (like our hidden Permutation task), while still keeping the scoring 100% deterministic so we don't have to rely on "LLM-as-a-judge." SkyDiscover is already using this for 200+ task benchmark suites to get much cleaner, less noisy data on agent performance."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Evaluating Evolving Agents with Evolving Benchmarks"},"story_url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["frontiercs"],"value":"https://frontier-cs.org/blog/agent-evaluation/"}},"_tags":["comment","author_lihanc111","story_47328312"],"author":"lihanc111","comment_text":"Evolving agent systems (agents that use continual learning, memory, or context refinement) are becoming the mainstream. However, up until now, most papers show gains on only about 10 tasks, or small subsets of existing benchmarks.
The bigger issue is benchmark saturation. We highlight "circle packing" in the post as a prime example: four different recent papers (ThetaEvolve, TTT-Discover, AdaEvolve, and EvoX) all converge to the exact same ceiling (2.635983). Once multiple methods collapse to the same endpoint because they are essentially just wrapping the same SLSQP Python script, the benchmark stops measuring open-ended improvement.
To fix the evaluation gap, we built Frontier-CS. It consists of 172 algorithmic tasks and 68 research tasks, designed by ICPC World Finalists and CS PhDs.
Our main goal was to create environments with massive search spaces that require actual information-efficient exploration and long-horizon reasoning (like our hidden Permutation task), while still keeping the scoring 100% deterministic so we don't have to rely on "LLM-as-a-judge." SkyDiscover is already using this for 200+ task benchmark suites to get much cleaner, less noisy data on agent performance.","created_at":"2026-03-10T20:19:41Z","created_at_i":1773173981,"objectID":"47328313","parent_id":47328312,"story_id":47328312,"story_title":"Evaluating Evolving Agents with Evolving Benchmarks","story_url":"https://frontier-cs.org/blog/agent-evaluation/","updated_at":"2026-03-10T20:27:44Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"skejriwal44"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["frontiercs"],"value":"We are open-sourcing SkyDiscover, a modular framework for LLM-driven algorithm discovery.
Most prior systems (e.g., AlphaEvolve) are closed-source. Existing open alternatives are monolithic and hard to extend. SkyDiscover decomposes discovery into four swappable components: Context Builder, Solution Generator, Evaluator, and Selector.
On top of this, we implemented two adaptive algorithms (~2.5K LoC each):
AdaEvolve \u2013 adapts search based on real-time progress
EvoX \u2013 allows the model to modify its own discovery strategy
Across 200+ benchmarks (math, systems, Frontier-CS, multimodal):
+34% median improvement on 172 Frontier-CS tasks over prior open methods
Matched/exceeded AlphaEvolve on several math and systems tasks
41% lower cross-cloud transfer cost
14% better MoE load balance (9.5\u00d7 faster)
29% lower KV-cache pressure vs. published human baseline
The goal is to make discovery algorithms easy to build, compare, and extend.
Feedback welcome."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"[dead]"}},"_tags":["comment","author_skejriwal44","story_47236656"],"author":"skejriwal44","comment_text":"We are open-sourcing SkyDiscover, a modular framework for LLM-driven algorithm discovery.
Most prior systems (e.g., AlphaEvolve) are closed-source. Existing open alternatives are monolithic and hard to extend. SkyDiscover decomposes discovery into four swappable components: Context Builder, Solution Generator, Evaluator, and Selector.
On top of this, we implemented two adaptive algorithms (~2.5K LoC each):
AdaEvolve \u2013 adapts search based on real-time progress
EvoX \u2013 allows the model to modify its own discovery strategy
Across 200+ benchmarks (math, systems, Frontier-CS, multimodal):
+34% median improvement on 172 Frontier-CS tasks over prior open methods
Matched/exceeded AlphaEvolve on several math and systems tasks
41% lower cross-cloud transfer cost
14% better MoE load balance (9.5\u00d7 faster)
29% lower KV-cache pressure vs. published human baseline
The goal is to make discovery algorithms easy to build, compare, and extend.
Feedback welcome.","created_at":"2026-03-03T18:32:26Z","created_at_i":1772562746,"objectID":"47236657","parent_id":47236656,"story_id":47236656,"story_title":"[dead]","updated_at":"2026-03-05T23:40:48Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"belter"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["frontiercs"],"value":""... a benchmark of 156 open-ended problems across diverse areas of computer science, designed and reviewed by experts, including CS PhDs and top-tier competitive programming participants and problem setters. Unlike existing benchmarks that focus on tasks with known optimal solutions, FrontierCS targets problems where the optimal solution is unknown, but the quality of a solution can be objectively evaluated...""},"story_title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["frontiercs"],"value":"FrontierCS: Evolving Challenges for Evolving Intelligence"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://arxiv.org/abs/2512.15699"}},"_tags":["comment","author_belter","story_46314520"],"author":"belter","comment_text":""... a benchmark of 156 open-ended problems across diverse areas of computer science, designed and reviewed by experts, including CS PhDs and top-tier competitive programming participants and problem setters. Unlike existing benchmarks that focus on tasks with known optimal solutions, FrontierCS targets problems where the optimal solution is unknown, but the quality of a solution can be objectively evaluated..."","created_at":"2025-12-18T16:30:35Z","created_at_i":1766075435,"objectID":"46314859","parent_id":46314520,"story_id":46314520,"story_title":"FrontierCS: Evolving Challenges for Evolving Intelligence","story_url":"https://arxiv.org/abs/2512.15699","updated_at":"2026-03-05T23:13:39Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jimsimmons"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["frontiercs"],"value":"Palantir is Infosys for governments. This remark is a psyop by Karp to make them seem like some frontier CS place."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Palantir CEO Rejects Calls to Pause AI Development"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://newsnotfound.com/palantir-ceo-rejects-calls-to-pause-ai-development-amid-global-debate/"}},"_tags":["comment","author_jimsimmons","story_36271120"],"author":"jimsimmons","comment_text":"Palantir is Infosys for governments. This remark is a psyop by Karp to make them seem like some frontier CS place.","created_at":"2023-06-10T16:34:29Z","created_at_i":1686414869,"objectID":"36271981","parent_id":36271120,"story_id":36271120,"story_title":"Palantir CEO Rejects Calls to Pause AI Development","story_url":"https://newsnotfound.com/palantir-ceo-rejects-calls-to-pause-ai-development-amid-global-debate/","updated_at":"2024-09-20T14:17:19Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"lihanc111"},"title":{"matchLevel":"none","matchedWords":[],"value":"Evaluating Evolving Agents with Evolving Benchmarks"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["frontiercs"],"value":"https://frontier-cs.org/blog/agent-evaluation/"}},"_tags":["story","author_lihanc111","story_47328312"],"author":"lihanc111","children":[47328313],"created_at":"2026-03-10T20:19:41Z","created_at_i":1773173981,"num_comments":1,"objectID":"47328312","points":3,"story_id":47328312,"title":"Evaluating Evolving Agents with Evolving Benchmarks","updated_at":"2026-03-10T23:00:30Z","url":"https://frontier-cs.org/blog/agent-evaluation/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"matt_d"},"title":{"matchLevel":"none","matchedWords":[],"value":"FrontierSmith: Synthesizing Open-Ended Coding Problems at Scale"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["frontiercs"],"value":"https://frontier-cs.org/blog/frontiersmith/"}},"_tags":["story","author_matt_d","story_48154112"],"author":"matt_d","created_at":"2026-05-15T21:27:28Z","created_at_i":1778880448,"num_comments":0,"objectID":"48154112","points":2,"story_id":48154112,"title":"FrontierSmith: Synthesizing Open-Ended Coding Problems at Scale","updated_at":"2026-05-16T01:43:59Z","url":"https://frontier-cs.org/blog/frontiersmith/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"matt_d"},"title":{"matchLevel":"none","matchedWords":[],"value":"Evaluating the Hardest CS Problems in the Age of LLMs"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["frontiercs"],"value":"https://frontier-cs.org/blog/evaluation/"}},"_tags":["story","author_matt_d","story_47082469"],"author":"matt_d","created_at":"2026-02-20T01:24:35Z","created_at_i":1771550675,"num_comments":0,"objectID":"47082469","points":1,"story_id":47082469,"title":"Evaluating the Hardest CS Problems in the Age of LLMs","updated_at":"2026-03-05T23:36:08Z","url":"https://frontier-cs.org/blog/evaluation/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"streamer45"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["frontiercs"],"value":"FrontierCode: An eval to measure whether you would actually merge the code"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://cognition.ai/blog/frontier-code"}},"_tags":["story","author_streamer45","story_48451723"],"author":"streamer45","children":[48452470,48452593,48452647,48452995,48453013,48453609,48455594,48455646,48456137,48456270,48456401,48457001,48457061,48458006,48458269,48458845,48459572,48460387],"created_at":"2026-06-08T20:45:03Z","created_at_i":1780951503,"num_comments":57,"objectID":"48451723","points":256,"story_id":48451723,"title":"FrontierCode: An eval to measure whether you would actually merge the code","updated_at":"2026-08-07T09:34:41Z","url":"https://cognition.ai/blog/frontier-code"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"meetpateltech"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["frontiercs"],"value":"Pushing the frontiers of audio generation"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["frontiercs"],"value":"https://deepmind.google/discover/blog/pushing-the-frontiers-of-audio-generation/"}},"_tags":["story","author_meetpateltech","story_41995730"],"author":"meetpateltech","children":[41996371,41996645,41997270,41997291,41998187,41998241,41998284,41998290,41999078,41999162,41999410,41999441,42000527,42003061,42010052],"created_at":"2024-10-30T15:02:56Z","created_at_i":1730300576,"num_comments":107,"objectID":"41995730","points":237,"story_id":41995730,"title":"Pushing the frontiers of audio generation","updated_at":"2024-11-11T21:10:36Z","url":"https://deepmind.google/discover/blog/pushing-the-frontiers-of-audio-generation/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"tambourine_man"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["frontiercs"],"value":"The Staggering Frontiers of CSS"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["frontiercs"],"value":"https://buttondown.email/robinrendle/archive/the-many-frontiers-of-css/"}},"_tags":["story","author_tambourine_man","story_36375179"],"author":"tambourine_man","children":[36375654,36375678,36375866,36375874,36376444,36376639,36378001,36378292,36378334,36378611],"created_at":"2023-06-17T22:06:38Z","created_at_i":1687039598,"num_comments":43,"objectID":"36375179","points":145,"story_id":36375179,"title":"The Staggering Frontiers of CSS","updated_at":"2026-07-29T22:46:34Z","url":"https://buttondown.email/robinrendle/archive/the-many-frontiers-of-css/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"imartin2k"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["frontiercs"],"value":"Don't Be Evil: Utopias, Frontiers, and Brogrammers"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://logicmag.io/03-dont-be-evil/"}},"_tags":["story","author_imartin2k","story_16065098"],"author":"imartin2k","children":[16067106,16067898,16068030,16068128,16068605,16068713,16068890,16068972,16069177,16069232,16071923],"created_at":"2018-01-03T21:01:13Z","created_at_i":1515013273,"num_comments":31,"objectID":"16065098","points":137,"story_id":16065098,"title":"Don't Be Evil: Utopias, Frontiers, and Brogrammers","updated_at":"2024-09-20T01:54:20Z","url":"https://logicmag.io/03-dont-be-evil/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"gopiandcode"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["frontiercs"],"value":"How to (actually) prove it \u2013 New Frontiers of Mathematics and Computing in Lean"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://kirancodes.me/posts/log-how-to-prove-it-maths.html"}},"_tags":["story","author_gopiandcode","story_43932871"],"author":"gopiandcode","children":[43956123,43956435,43958932,43961156,43963525],"created_at":"2025-05-09T01:00:53Z","created_at_i":1746752453,"num_comments":17,"objectID":"43932871","points":81,"story_id":43932871,"title":"How to (actually) prove it \u2013 New Frontiers of Mathematics and Computing in Lean","updated_at":"2026-08-04T14:41:16Z","url":"https://kirancodes.me/posts/log-how-to-prove-it-maths.html"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"louiskirsch"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["frontiercs"],"value":"Reinforcement Learning \u2013 The different challenges, methods and frontiers"},"url":{"matchLevel":"none","matchedWords":[],"value":"http://louiskirsch.com/maps/reinforcement-learning"}},"_tags":["story","author_louiskirsch","story_18830578"],"author":"louiskirsch","children":[18836522],"created_at":"2019-01-05T08:27:27Z","created_at_i":1546676847,"num_comments":1,"objectID":"18830578","points":69,"story_id":18830578,"title":"Reinforcement Learning \u2013 The different challenges, methods and frontiers","updated_at":"2024-09-20T03:30:01Z","url":"http://louiskirsch.com/maps/reinforcement-learning"}],"hitsPerPage":20,"nbHits":33041,"nbPages":50,"page":0,"params":"query=FrontierCS&advancedSyntax=true&analyticsTags=backend","processingTimeMS":10,"processingTimingsMS":{"_request":{"queue":28,"roundTrip":23},"fetch":{"query":5,"scanning":3,"total":9},"total":10},"query":"FrontierCS","serverTimeMS":39}