En 30 secondes
In March 2023, AI Explained compares early Google Bard responses with GPT-4-powered Bing using selected prompts. The reviewer reports errors in both, differences in writing and image prompts, and a theory-of-mind test that should not be taken as proof of model understanding.
- 1Local searchBoth assistants miss a walking-distance constraint in the example.
- 2Math and datesThe reviewer finds errors in calculations and dates.
- 3Creative writingThe author’s preferences differ by task and are subjective.
- 4Image promptsBard produces image prompts the reviewer prefers in one test.
- 5InterpretationSelected answers do not prove a model possesses human-like understanding.
À retenir
- Both assistants miss a walking-distance constraint in the example.
- The reviewer finds errors in calculations and dates.
- The author’s preferences differ by task and are subjective.
- Bard produces image prompts the reviewer prefers in one test.
- Selected answers do not prove a model possesses human-like understanding.
Lire le texte remanié, chapitre par chapitre
Early search and math tests
The reviewer compares selected Bard and Bing outputs.
In March 2023, shortly after gaining access to Google Bard, the reviewer compares selected responses with Bing powered by GPT-4. These are informal early tests, not a controlled benchmark or an assessment of current products.
Both assistants struggle with a local-search question asking for florists within a ten-minute walk of the British Museum. The reviewer says the returned locations do not meet the requested walking constraint.
Bard gives incorrect or misleading answers to a percentage problem and an algebra example in these tests, while Bing succeeds on the examples shown. Different Bard drafts do not consistently fix the errors.
Both systems get a question about the interval between historical events wrong. Bard apologizes when challenged. The reviewer’s lesson is to verify dates and calculations rather than relying on fluent explanations.
Writing and image prompts
Results differ across creative tasks.
In a subjective joke-generation comparison, the reviewer prefers some of Bard’s shorter jokes, even though he does not find them particularly funny. Humor quality is a matter of judgment.
A grammar-correction example favors Bing in the reviewer’s trials. He also prefers Bing’s sonnet about modern London, finding Bard’s version less vivid. These judgments apply to the prompts and outputs shown.
He notes a speed and scale tradeoff: Bard appeared faster in his early experience. His explanation of underlying model size and serving choices is a conjecture, not a verified architectural comparison.
When asked to propose Midjourney image prompts, Bard produces examples the reviewer prefers. He contrasts this with Bing’s prompts and notes that prompt wording and iteration affect the result.
Jokes and interpretation
The reviewer tests explanation and a theory-of-mind prompt.
The reviewer tests whether the systems can explain a series of jokes. Both handle several examples; in one joke about living in the same house and family as one’s parents, he finds Bing recognizes the intended humor while Bard responds literally.
A final prompt asks what the user thinks about the model’s theory of mind. The reviewer prefers Bing’s response, but a convincing answer to a selected question does not establish that a system has human-like mental states or theory of mind.
The review closes with a reminder that the author has only just gained access and intends further testing. Capabilities, product names, interfaces and pricing may change after this 2023 snapshot.
Sur la même chaîne
31:00EN
AI Explained· 12 oct. 2026Why AI Labs Keep Racing Despite Their Own Warnings of a Major BreachSam Altman, John Schulman · OpenAIAI Explained examines weakening safeguards, the incentives behind the frontier race, and why research breakthroughs make the next 18 months difficult to forecast.
38:28ENAI Explained· 1 oct. 2026AI Agent Security and the Challenge of Recursive Self-ImprovementAI Explained examines reported containment failures, model evaluation and the pressure to accelerate AI research.
32:54ENAI Explained· 24 sept. 2026Opus 5.5 and the Race Toward Automated AI ResearchAI Explained weighs model progress against the increasingly difficult task of evaluating and governing recursive self-improvement.
24:53ENAI Explained· 16 sept. 2026Why Some Researchers Want to Pace Frontier AIAI Explained surveys possible scaling routes, monitoring concerns and the debate over how capabilities could lead to harm.
29:05ENAI Explained· 4 sept. 2026GPT-6 Astra’s Benchmark Gains and the Challenge of Monitoring Its ReasoningAI Explained examines reported gains across scientific, mathematical and agent benchmarks alongside concerns about less observable reasoning.
23:41ENAI Explained· 27 août 2026AI Agents, Unintended Training Rewards and the Challenge of OversightAI Explained examines reports on agent coordination, training incentives and the growing use of AI to evaluate other AI systems.