Alternatives
Products that do what Humor Arena – Which frontier model is funniest? does
What if you could measure humor? Well we've trained a model on our own dataset of ~50k human ratings to detect what jokes people find funniest. We know it's part objective, part subjective component. Subjective is out of our depth for now haha The main results: Fable 5 is funniest - beating the average model 67% of the time, with GPT 4o last at 17%. Other findings: - The models never refused to try, even with dark prompts - Thinking longer has a slight benefit - Absurdness correlates negatively with joke quality Some methodology notes: - We benchmarked our model against the human majority…
- 1
- 2MO
I wanted to share our new speech to text model, and the library to use them effectively. We're a small startup (six people, sub-$100k monthly GPU budget) so I'm proud of the work the team has done to create streaming STT models with lower word-error rates than OpenAI's largest Whisper model. Admittedly Large v3 is a couple of years old, but we're near the top the HF OpenASR leaderboard, even up against Nvidia's Parakeet family. Anyway, I'd love to get feedback on the models and software, and hear about what people might build with it.
Feb 2026 · github.com
- 3
- 4OA
Hi HN, we built world-model-optimizer, an open source tool to continually improve a specialized model for an agent. It does this by simulating production tool responses through text world modeling (similar to QwenAgentWorld, summary here https://x.com/silennai/status/2073887455884058814). We can then use this to train a router for frontier, OS, and local models (use defaults or pick which ones to optimize against). wmo ingests agent traces, builds the simulation, embeds the traces, runs different models you choose against the simulation scenarios, and then uses a KNN…
Jul 2026 · github.com
- 5FT
Aug 2026 · github.com
- 6WF
We have a dataset of 3,095 standardized AI responses across 43 prompts. From each response, we extract a 32-dimension stylometric fingerprint (lexical richness, sentence structure, punctuation habits, formatting patterns, discourse markers). Some findings: - 9 clone clusters (>90% cosine similarity on z-normalized feature vectors) - Mistral Large 2 and Large 3 2512 score 84.8% on a composite metric combining 5 independent signals - Gemini 2.5 Flash Lite writes 78% like Claude 3 Opus. Costs 185x less - Meta has the strongest provider "house style" (37.5x distinctiveness ratio) - "Satirical…
Apr 2026 · rival.tips
- 7
Track how AI models feel in everyday use through public community feedback, 7-day experience scores and trends. This is not a capability benchmark.
23d ago · isaidumber.today
- 8
- 9RR
Hey HN - I'm Mason, and I'm excited to share a fun side project I've been working on recently: Repo Roast. For me, Github often feels like a serious (and sometimes soulless) place. I built this website to showcase the human, silly side of developers. Repo Roast is built on top of LLM-powered GitHub analytics tooling we’re developing internally at Merit Systems [0]. One of the historically cited weak points for LLMs has been their sense of humor. I found that with the right prompt, at scale, they mostly can _identify_ humor pretty well. We experimented with a couple of different models to…
2025 · reporoast.fun
- 10VI
Most inference UIs that I've come across pretty much just give us a chat-like interface to toy around with models in a single visual conversation thread. Given the fact that we are limited to seeing only one output at a time, it's kind of hard to compare outputs from different models, adjustments made to the prompting, and sampler settings. But even when keeping the generation parameters the same (e.g., to test for reliability in the output) and just going for multiple passes, there is no easy way to have a side-by-side comparison to keep track of the outputs from the multiple "rounds". I…
2024 · github.com
- 11IR
The Emotion Engine has 32 MB of RAM total, so the trick is streaming weights from CD-ROM one matrix at a time during the forward pass — only activations, KV cache and embeddings live in RAM. This means models bigger than the RAM can still run, they just read more from disc. Had to build a custom quantized format (PSNT), hack endianness, write a tokenizer pipeline, and most of the PS2 SDK from scratch (releasing that separately). The model itself is also custom — a 10M param Llama-style architecture I trained specifically for this. And it works. On real hardware.
Mar 2026 · github.com
- 12L2
Hi all, today we're excited to launch LoraLand: 25 fine-tuned mistral-7b models that outperform #gpt4 on task-specific applications ranging from sentiment detection to question answering. All 25 fine-tuned models… - Outperform GPT-4, GPT-3.5-turbo, and mistral-7b-instruct for specific tasks - Are cost-effectively served from a single GPU through LoRAX - Were trained for less than $8 each on average You can prompt all of the fine-tuned models today and compare their results to mistral-7b-instruct in real time! We'd love to hear comments and feedback from the community
2024 · predibase.com
- 13SO
I’ve been building a crowd-sourced AI detection benchmark. Two responses to the same prompt — one from a real human (pre-2022, provably pre prevalence of AI slop on the internet), one generated by AI. You pick the slop. Three wrong and you’re out. The dataset: 16K human posts from Reddit, Hacker News, and Yelp, each paired with AI generations from 6 models across two providers (Anthropic and OpenAI) at three capability tiers. Same prompt, length-matched, no adversarial coaching — just the model’s natural voice with platform context. Every vote is logged with model, tier, source, response…
Mar 2026 · slop-or-not.space
- 14
- 15CB
Hey HN, we're excited to share Cua-Bench ( https://github.com/trycua/cua ), an open-source framework for evaluating and training computer-use agents across different environments. Computer-use agents show massive performance variance across different UIs—an agent with 90% success on Windows 11 might drop to 9% on Windows XP for the same task. The problem is OS themes, browser versions, and UI variations that existing benchmarks don't capture. The existing benchmarks (OSWorld, Windows Agent Arena, AndroidWorld) were great but operated in silos—different harnesses,…
Jan 2026 · github.com
- 16FT
Dec 2025 · nixiesearch.substack.com
- 17ΤB
τ-Bench is an open benchmark for evaluating AI agents on grounded, multi-turn customer service tasks with verifiable outcomes. It's been great to see the community adopt it since launch — this is now the third iteration. With τ³-Bench, we're extending it to two new settings: knowledge-intensive retrieval and full-duplex voice. τ-Knowledge: agents must navigate ~700 interconnected policy documents to complete multi-step tasks. Best frontier model (GPT-5.2, high reasoning) hits ~25%. The surprising part: even when you hand the model the exact documents it needs, performance only reaches ~40%.…
Mar 2026
- 18WM
We wanted to test if a smaller model like GPT-4.1-mini could beat its bigger brother 4.1 at the game Tic-Tac-Toe using only context engineering. We put them in a 100-game tournament. For the smaller model, we gave it a few examples of winning moves from past games right before it made its own move. The results were clear. Without the examples, the smaller model struggled against GPT-4.1. With the examples, its effectiveness increased by nearly 200%, and it consistently won. It's a simple demonstration, but it shows that a smaller, faster model with good, timely examples can outperform a more…
2025 · github.com
- 19PG
I’m Andrew, co-founder of Recall. Over the past few days I’ve been building Predict, a playground where anyone can: - propose skills we should measure in language models—live examples include difficult math, memory-manipulation resistance, code generation, and empathy under bad news - write evals (graded prompts) for those skills - forecast which models will score highest once GPT-5 is released Why this exists Benchmarks leak into training data quickly; scores are unreliable and labs still declare progress. The prediction tool aims keeps the target moving by letting the crowd define both the…
2025
- 20GV
2024 · canyouexplainthejoke.com
- 21NL
Refuel LLM (84.2%) outperforms trained human annotators (80.4%), GPT-3-5-turbo (81.3%), PaLM-2 (82.3%) and Claude (79.3%) across a benchmark of 15 text labeling datasets. It is a Llama-v2-13b base model, trained on over 2500 unique datasets (5.24B tokens) spanning categories such as classification, entity resolution, matching, reading comprehension and information extraction. Here is the interactive demo: https://labs.refuel.ai/playground. Pretty fun to play with!
2023
- 22
- 23AM
Hey, HN community! I'm excited to share the fifth issue of our AI/ML Weekly Digest. This innovative newsletter uses the power of GPT-4 to analyse and curate the most relevant and exciting AI/ML stories from Hacker News. This week I also share with our subscribers a curated list of resources during my learning journey https://github.com/vlameiras/ai-ml-resources/ GPT-4 scours through the top stories on Hacker News to bring you a concise summary and sentiment analysis of the hottest AI/ML news each week. Subscribe & Stay Updated To get the complete…
2023 · hn-ai-newsletter.beehiiv.com
- 24AS
I'm a combat veteran living paycheck to paycheck with no computer science degree. I built an AI system that benchmarks 60x faster than industry leaders. Real benchmarks (Dec 12, 2025): - 3.43ms response time (vs 50-200ms industry average) - 337 queries/second (vs 50-150) - 0% error rate, 100% uptime - Constitutional AI with 1,235 specialized "brains" Built it in 3 weeks. 4 U.S. patents pending. Full story + independent benchmarks: https://thebrokenwayfoundation.org Not asking for money. Just need technical validators to verify this is real.
Dec 2025
Ranked by how close each launch is in meaning, then by votes. Refine with a description →