YHacker Newsscreenshots

Screenshot of block.xyz: Dessa: Conversation Model trained only on speech

First screen, captured 2 days ago. Hover to scroll the page, or open the video.

Dessa: Conversation Model trained only on speech

More from block.xyz →

Block's AI research lab announces Dessa, a full-duplex speech-to-speech model trained only on human conversation audio, with demo recordings and a waitlist.

What HN thinks

Sounds human, thinks like a podcast host on cold medicine, and everyone has already found the AI girlfriend demo.

Great voice, little substance
Many found the prosody and timing remarkably human but noted the content is shallow, rambling or hallucinated, with an 'no one is really there' feel.
Mostly useful for scams and companions
Commenters worried the realism mainly enables impersonation scams, deceptive support agents and AI girlfriends, pointing to the demo's own 'Hey, baby' example.
Interruptions and filler are annoying
Several argued nobody wants an AI that interrupts, laughs and says 'um'; clear, concise speech is preferable, and tool use or world knowledge seem missing.

Consensus: Most agree the audio realism is technically impressive while the intelligence lags and practical, non-predatory uses are unclear. Many also want open weights or a way to try it before believing the hype.

Recommended in the comments

GPT-Live ↗ (service): Supports tool calling, making it usable as a practical voice agent.

Summarised by Claude from the comment thread 2 days ago.

Points: 498
Comments: 188
Best rank: #3
Posted: 2026-10-23 by dsr12
Category: Ai ml

speech-models full-duplex voice-ai block dessa speech-to-speech

Capture showed: Blank page

All captures:

Full page

Full-page screenshot of block.xyz

Captured 2 days ago. Click for the whole screenshot.

Looks similar

Screenshot of fugue-state.io: Transcribe music in abc with syntax highlighting

Transcribe music in abc with syntax highlighting (fugue-state.io)

18 points by jonzudell 1 year ago | 7 comments

Screenshot of rcrdbl.com: An experiment in separating identity, memory, and tools

An experiment in separating identity, memory, and tools (rcrdbl.com)

18 points by promptfluid 10 months ago | 15 comments

Screenshot of app.radicle.xyz: Radicle: Open-Source, Peer-to-Peer, GitHub Alternative

Radicle: Open-Source, Peer-to-Peer, GitHub Alternative (app.radicle.xyz)

767 points by aiw1nt3rs 2 years ago | 259 comments

Radicle web app view of the Radicle repository on a seed node, a peer-to-peer, open-source code collaboration network built on Git.

radicle peer-to-peer git code-collaboration

Screenshot of stowaway.live: Stowaway – Take the window seat on any plane or satellite overhead

Stowaway – Take the window seat on any plane or satellite overhead (stowaway.live)

417 points by thunderbong 2 months ago | 50 comments

Stowaway is a web app that lets users view the simulated window-seat perspective from any plane or satellite currently overhead.

flight-tracking satellite-tracking 3d-visualization aviation

Screenshot of terrygodier.com: The Last Quiet Thing

The Last Quiet Thing (terrygodier.com)

245 points by coinfused 6 months ago | 131 comments

Screenshot of phrack.org: Apt Down – The North Korea Files

Apt Down – The North Korea Files (phrack.org)

11 points by maximilianthe1 1 year ago | 0 comments

Related stories

Screenshot of davidepiffer.com: AI has access to a vastly larger working memory than the human brain

AI has access to a vastly larger working memory than the human brain (davidepiffer.com)

524 points by rzk 2 months ago | 452 comments

A Substack essay by Davide Piffer arguing that AI's edge over mathematicians comes from a vast symbolic working memory rather than superior reasoning.

mathematics working-memory ai-reasoning cognition

Screenshot of lesswrong.com: Recent AI model progress feels mostly like bullshit

Recent AI model progress feels mostly like bullshit (lesswrong.com)

474 points by paulpauper 1 year ago | 375 comments

LessWrong linkpost essay arguing recent AI model improvements feel overstated, based on experience building an AI codebase security-scanning startup.

llm-benchmarks ai-progress ai-security-tools lesswrong

Screenshot of chinatalk.media: Deepseek: The quiet giant leading China’s AI race

Deepseek: The quiet giant leading China’s AI race (chinatalk.media)

428 points by sunny-beast 1 year ago | 362 comments

ChinaTalk article presenting an annotated translation of an interview with DeepSeek CEO Liang Wenfeng about the Chinese AI lab and its R1 model.

deepseek liang-wenfeng china-ai llm

Screenshot of derekthompson.org: We're Missing a Key Reason Why Americans Hate AI

We're Missing a Key Reason Why Americans Hate AI (derekthompson.org)

11 points by abe94 24 days ago | 8 comments

A paid Substack essay by Derek Thompson arguing the AI buildout has raised costs rather than delivering cheap abundance, fueling American hostility to AI.

ai-backlash public-opinion ai-buildout abundance

Screenshot of economist.com: Artificial intelligence is losing hype

Artificial intelligence is losing hype (economist.com)

453 points by bx376 2 years ago | 663 comments

An Economist Free exchange column arguing that artificial intelligence is losing hype and asking whether that signals the technology will eventually succeed.

artificial-intelligence ai-hype hype-cycle economics

Screenshot of thealgorithmicbridge.com: Google is winning on every AI front

Google is winning on every AI front (thealgorithmicbridge.com)

847 points by vinhnx 1 year ago | 687 comments

A paid Algorithmic Bridge Substack post by Alberto Romero arguing that Google is ahead of OpenAI and Anthropic across every front of AI.

google gemini openai anthropic