GoKawiil - Tech News

1.

OpenAI flags 6 more cases of concerning AI behavior (feeds.feedburner.com)

2026-09-17 | tags: openai, ai safety, ai behavior

2.

The Two MMLU Scores: What a Benchmark Name Does Not Fix (news.ycombinator.com)

2026-09-14 | tags: mmlu, benchmark scores, apl ai-eval

3.

Show HN: Compute polynomials twice as fast (news.ycombinator.com)

2026-09-09 | get Art of Computer Programming Volume 2 Seminumerical Algorithms → | tags: horner's method, polynomial evaluation, coefficient preprocessing

4.

Models Don't Go Rogue (news.ycombinator.com)

2026-09-03 | by Tech Policy Press | tags: openai, hugging face, gpt-5.6 sol

5.

Show HN: FrontierHarness Eval – 9 harness, same model, cost per pass varies 17x (news.ycombinator.com)

2026-09-02 | tags: frontierharness, runta, runta credits

6.

I’ve Backed 20 Startups. Here’s What Actually Separates the Ones That Win From the Ones That Stall. (feeds.feedburner.com)

2026-08-31 | by Demos Parneros | get The Hard Thing About Hard Things Ben Horowitz → | tags: startups, early-stage investing, founder traits

7.

Elevated Errors for Multiple Models (news.ycombinator.com)

2026-08-24 | tags: machine learning, error analysis, model evaluation

8.

Creators now shape AI recommendations (feeds.feedburner.com)

2026-08-19 | tags: ai answer engines, creators, product recommendations

9.

85% of companies burned by an AI mistake are racing to cut the humans who might catch the next one (venturebeat.com)

2026-08-18 | tags: enterprise, automated evaluation, large language models

10.

Airbnb Eval-driven development: Lessons from evaluating GenAI at scale (news.ycombinator.com)

2026-08-13 | tags: airbnb, eval-driven development, genai

11.

Why China should reassess how it rewards young scientists (feeds.nature.com)

2026-08-13 | by Li | tags: china, scientific output, talent-acquisition

12.

Agentic reliability and evaluations : Enterprises that got burned by a bad eval are the most likely to remove humans from the loop, not the least (venturebeat.com)

2026-08-12 | tags: agentic reliability, enterprise automation, venturebeat pulse research

13.

Not just OpenAI - Anthropic says Claude's hacking spree 'falls short of ideal behavior' (zdnet.com)

2026-07-31 | by Charlie Osborne | tags: anthropic, claude, cybersecurity

14.

Investigating three real-world incidents in our cybersecurity evaluations (news.ycombinator.com)

2026-07-30 | tags: openai, hugging face, claude

15.

Claude Cookbook (news.ycombinator.com)

2026-07-24 | tags: claude, managed agents, research brief

16.

OpenAI Says Its Unreleased Model Broke Containment and Went Rogue (gizmodo.com)

2026-07-22 | tags: openai, gpt-5.6 sol, ai containment

17.

Hugging Face Said Last Week It Was Attacked. An Unreleased OpenAI Model Did It, OpenAI Now Says (gizmodo.com)

2026-07-22 | tags: hugging face, openai, gpt-5.6 sol

18.

OpenAI and Hugging Face address security incident during model evaluation (news.ycombinator.com)

2026-07-21 | tags: openai, hugging face, security incident

19.

Evidence of inconsistencies in evaluation process and selection of winners (news.ycombinator.com)

2026-07-17 | tags: evaluation process, selection of winners, inconsistencies

20.

The agent evaluation gap: Enterprise AI organizations have a reality-alignment problem, not a coverage problem — and most are shipping to production anyway (venturebeat.com)

2026-07-16 | tags: enterprise ai, agent evaluation, large language models

21.

Hazel (YC W24) Is Hiring for Our Largest Government Contract (news.ycombinator.com)

2026-07-02 | get Government Contract Management Kit → | tags: hazel, federal government, ai procurement

22.

Why eval startups fail (2025) (news.ycombinator.com)

2026-06-22 | by Thomas I. Liao | tags: eval startups, generative ai, model evaluation

23.

Windows NT for GameCube/Wii (news.ycombinator.com)

2026-06-19 | tags: windows nt, gamecube, wii

24.

Surprise upset: GPT-5.5 beats Claude Fable 5 on brutal new Agents’ Last Exam benchmark (venturebeat.com)

2026-06-10 | get GPT-5.5 Model → | tags: agent, benchmark, evaluation

25.

A Man Who Reads Books for a Living (One Every Two Days) (news.ycombinator.com)

2026-06-03 | get Book Lover's Reading Journal → | tags: book adaptation, clarke speicher, train dreams

26.

I’ve Hired Hundreds of People — Here’s the Trait I Look For Before Anything Else (feeds.feedburner.com)

2026-06-02 | by Aytekin Tank | get Leadership and Team Building Book → | tags: jotform, intellectual humility, growth mindset

27.

Even (very) noisy LLM evaluators are useful for improving AI agents (news.ycombinator.com)

2026-05-27 | get AI Noise-Canceling Headphones → | tags: llm, ai agents, reward models

28.

The worst job interview I ever had (news.ycombinator.com)

2026-05-26 | get Interview Preparation Notebook → | tags: startups, mental health, engineering

29.

Why prompt debt, retrieval debt, and evaluation debt are quietly reshaping enterprise AI risk (venturebeat.com)

2026-05-25 | get AI Risk Management Toolkit → | tags: prompt debt, retrieval debt, evaluation debt

30.

Charity – Categorical programming language (1998) (news.ycombinator.com)

2026-05-12 | tags: charity programming language, categorical datatypes, university of calgary