/
Navigation
Chronicles
Browse all articles
Explore
Semantic exploration
Research
Entity momentum
Nexus
Correlations & relationships
Story Arc
Topic evolution
Drift Map
Semantic trajectory animation
Posts
Analysis & commentary
Pulse API
Tech news intelligence API
Browse
Entities
Companies, people, products, technologies
Domains
Browse by publication source
Handles
Browse by social media handle
Detection
Concept Search
Semantic similarity search
High Impact Stories
Top coverage by position
Sentiment Analysis
Positive/negative coverage
Anomaly Detection
Unusual coverage patterns
Analysis
Rivalry Report
Compare two entities head-to-head
Semantic Pivots
Narrative discontinuities
Crisis Response
Event recovery patterns
Connected
Search: /
Command: ⌘K
Embeddings: large
TEXXR

Chronicles

The story behind the story

days · browse · Enter similar · o open

In an email, Google's Search SVP tells staff to rewrite Bard's bad responses in a “neutral tone” and to “not imply emotion” as the tool “learns best by example”

Jennifer Elias / CNBC :

CNBC Jennifer Elias

Context & Ripple Effects

Google’s instruction follows internal backlash to Bard’s unveiling, with employees calling the launch botched and un-Googley. It turns that criticism into an operational task: staff are being asked to supply model examples rather than merely flag failures.

The directive also sits alongside earlier reporting that Google’s AI organization exerted tighter control over work on sensitive AI topics, making the wording of Bard’s answers a product-governance issue as well as a training one.

First-order effects

  • Google staff reviewing Bard responses must rewrite poor outputs in a neutral, non-emotive style, creating curated examples for the chatbot to learn from.
  • Bard’s near-term feedback loop is steered toward Google-approved tone and framing, rather than preserving the language of its flawed answers.

Second-order effects

  • Employee criticism of the Bard launch gains a concrete channel into the product process, while also increasing the burden on Google’s reviewers to define acceptable responses consistently.
  • Google’s leadership takes a more direct role in setting conversational behavior, narrowing the distinction between model-quality work and communications control.

Third-order effects

  • If this review pattern persists, consumer AI assistants will be differentiated not only by underlying models but by the human editorial policies embedded in their training examples.
  • The episode points to a durable governance tension at Google: accelerating public AI releases while centrally controlling how sensitive or consequential answers are expressed.

The trend: Generative-AI companies are turning internal human review into a core mechanism for shaping both model quality and the acceptable tone of public-facing answers.

Discussion

  • @hughlangley Hugh Langley on x
    Scoop: Sundar Pichai is asking Googlers to each contribute 2-4 hours to train its Bard chatbot before it goes public. “AI has gone through many winters and springs. And now it is blooming again.” https://www.businessinsider.com/ ...
  • @ichetan @ichetan on x
    @Techmeme @jenn_elias Google is likely to fuck up in its incremental releases as they march into the search battlefield which has evolved; their opponent has an awesome combat and is determined to disrupt the search experience forver, a lesson learnt by trailing for 2 decades.
  • @alexandr_wang Alexandr Wang on x
    Google is offering an internal badge to their employees who help RLHF Bard. And slowly, RLHF will eat the world 😋 https://www.cnbc.com/...
  • @jenn_elias Jennifer Elias on x
    Trying to push Bard forward, Google execs are asking employees to rewrite its bad responses, saying it “learns best by example.” They've also sent “Do's and Don'ts” that tells them not to “imply emotion or claim to have human-like experiences.” https://www.cnbc.com/...