Consensus

AI academic search engine

writingno codeFree tier

Overview

Consensus searches across 200M+ academic papers to answer research questions with AI-synthesized summaries backed by citations. Shows consensus meters (what % of studies agree). Free tier includes 20 searches/day. Premium adds GPT-4 summaries and unlimited searches.

Best For

+Academic research
+Evidence-based decisions
+Literature reviews
+Students

News & Updates

Brothers Launch 'The Bias' iOS App to Synthesize News Coverage, Fight Media Bias

Charlie and Sam, two brothers, launched The Bias, an iOS news app that synthesizes coverage from multiple outlets into a single write-up for each major event. The app aims to combat media bias by clearly showing what coverage consensus exists, what key disagreements are, and what facts are undisputed, without requiring users to open ten browser tabs. This provides a faster alternative to manually aggregating perspectives and challenges the single-outlet framing of traditional news feeds. An Android version is reportedly coming soon.

NEW_TOOLS3/4/2026

CollectivIQ crowdsources AI answers, shows ChatGPT, Gemini, Claude, Grok & 10 others at once

Startup CollectivIQ launched a platform that aggregates responses from up to 14 different AI models, including ChatGPT, Gemini, Claude, and Grok, for a single user query. The system displays all answers simultaneously, allowing users to compare outputs and identify consensus or outliers. This approach directly tackles the 'hallucination' problem by letting users cross-reference multiple sources instead of trusting a single black-box response.

BREAKTHROUGHS3/4/2026

Researchers Propose Tool Verification Method to Fix TTRL's 'Spurious Consensus' Problem

Researchers Ruotong Liao, Nikolai Röhrich, and Xiaohan Wang introduced a tool verification framework for test-time reinforcement learning (TTRL), which is used to make large reasoning models self-evolve on unlabeled data. Their method addresses a critical flaw where TTRL systems can form a 'spurious yet high-frequency unverified consensus' through majority voting, leading to incorrect self-rewards. This verification layer could prevent TTRL models from reinforcing their own mistakes during online adaptation, improving reliability for real-world deployment.

NEW_TOOLS3/3/2026

Hacker News Post Warns LLMs Have Permanently Eroded Digital Privacy Standards

A post on Hacker News titled 'LLMs killed the privacy star, we can't rewind, we've gone too far' argues that large language models have fundamentally and irreversibly damaged data privacy. The post, which received 1 point and 0 comments, suggests the scale of data ingestion by models like GPT-4 and Claude has crossed a point of no return for user confidentiality. This reflects a growing technical consensus that the architecture of modern AI makes traditional data anonymization and deletion nearly impossible.

other2/28/2026

Citrini Research AI Economics Note Criticized as Fundamentally Flawed

A research note from Citrini Research on AI economics was posted to Hacker News and received immediate criticism for getting its core economic analysis wrong. The note, which presumably made claims about AI's cost structure, ROI, or market impact, was downvoted to 1 point with zero supporting comments, indicating a consensus rejection of its premises. This public dismissal on a major tech forum highlights how quickly flawed AI economic models can be identified and sidelined by the informed developer community.

BREAKTHROUGHS2/27/2026

Loading AI Digest