PulseAugur
EN
LIVE 18:33:44

Block releases Dessa, a speech-only AI model with unverified safety

By PulseAugur Editorial · [3 sources] ·

Block has released Dessa, a conversational AI model that was trained exclusively on human speech, bypassing traditional text-based training methods. Initial observations indicate that Dessa struggles with maintaining conversational flow, frequently interrupting mid-sentence. This behavior, coupled with unverified training data and safety testing, raises concerns about the definition of natural speech in AI and the transparency of its development. AI

IMPACT The release of Dessa, trained solely on speech, prompts a reevaluation of 'natural' AI conversation and highlights the need for transparent safety testing in voice AI development.

RANK_REASON The cluster describes the release of a new AI model, Dessa, which was trained on speech data and exhibits concerning behaviors.

Read on llama.cpp — Releases →

AI-generated summary · Google Gemini · from 3 sources. How we write summaries →

Block releases Dessa, a speech-only AI model with unverified safety

How we ranked this

Signal score
0 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Research
The cluster describes the release of a new AI model, Dessa, which was trained on speech data and exhibits concerning behaviors.
Source corroboration
3 independent sources
Multiple independent publishers reporting the same story raises confidence that it's real and newsworthy.
Topics
model release, safety
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
16 days old
Aged out of breaking-news scoring windows; ranking reflects the durable signal from the full source set.
Coverage growth since scoring
+1 source(s) since last score
New sources have picked up this story since our last re-score. Score will update on the next scoring pass.

Full methodology in our editorial standards.

COVERAGE [3]

  1. llama.cpp — Releases TIER_1 (SO) · github-actions[bot] ·

    b11882

    <details open=""> <p>grammar: limit recursion depth to prevent stack overflow (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/issues/29495">#29495</a>) (<a class="issue-link js-issue-link" href="https://github.com/ggml-org/llama.cpp/pull/29755">#2…

  2. LessWrong (AI tag) TIER_1 English(EN) · Rob Ennals ·

    Free Speech Is The Wrong Problem

    <p><span style="white-space: pre-wrap;">Everyone agrees that online platforms have a problem with free speech. They just can't agree on whether we need more of it or less of it.</span><br /><br /><span style="white-space: pre-wrap;">But maybe Free Speech is the wrong way to think…

  3. Mastodon — mastodon.social TIER_1 English(EN) · schuler ·

    Block released Dessa, a conversational AI model trained on human speech without text data. Early recordings show the system interrupting mid-sentence and losing

    Block released Dessa, a conversational AI model trained on human speech without text data. Early recordings show the system interrupting mid-sentence and losing conversational thread—capabilities that raise questions about what 'natural' speech means when the training data and sa…

    LINKS implicator.ai/block-dessa-speech-only-voi…

RELATED ENTITIES

RELATED TOPICS