FoundersArvind Sontha FounderArvind Sontha FounderSimilar JobsVoicepanelFounding Account ExecutivePureFounding GTM Leader (Sales)Brainbase LabsFounding GTMCraniometrixSales Director (Series A Healthtech)Saphira AIFounding Account Executive - Enterprise Safety & Compliance (Hardware + AI)TractianSales Account ExecutiveNotabeneSales Director, Americas (Institutional)EmbraceBusiness Development LeadTerminalFounding Account Executive [In-Person, Toronto]KomboBusiness Development LeadAptenFounding SDRLandeedBusiness Development And Enterprise Sales (BDES)ScoutSales DevelopmentExaAccount ExecutiveZaymoMidmarket Customer Success ManagerEnerjazzSales Manager – Field & Telecalling Team LeadNewsCatcherSales Development Representative (SDR)Surface LabsSales Development RepresentativeMacadamiaFounding GTM Lead JuiceboxCustomer Success ManagerFooterY CombinatorMake something people want.ProgramsYC ProgramStartup SchoolWork at a StartupCo-Founder MatchingResourcesStartup DirectoryStartup LibraryInvestorsDemo DaySafeHacker NewsLaunch YCYC DealsCompanyYC BlogContactPressPeopleCareersPrivacy PolicyNotice at CollectionSecurityTerms of UseTwitterTwitterFacebookFacebookInstagramInstagramLinkedInLinkedInYoutubeYouTube© 2026 Y Combinator
Even though my dataset is very small, I think it's sufficient to conclude that LLMs can't consistently reason. Also their reasoning performance gets worse as the SAT instance grows, which may be due to the context window becoming too large as the model reasoning progresses, and it gets harder to remember original clauses at the top of the context. A friend of mine made an observation that how complex SAT instances are similar to working with many rules in large codebases. As we add more rules, it gets more and more likely for LLMs to forget some of them, which can be insidious. Of course that doesn't mean LLMs are useless. They can be definitely useful without being able to reason, but due to lack of reasoning, we can't just write down the rules and expect that LLMs will always follow them. For critical requirements there needs to be some other process in place to ensure that these are met.
。搜狗输入法下载对此有专业解读
Сайт Роскомнадзора атаковали18:00
有趣的是，报告通过构建「2028 年宏观假想模型」，详细拆解了这一死循环的传导路径。
Филолог заявил о массовой отмене обращения на «вы» с большой буквы09:36