re: The Hidden Danger of Using AI for Answers

quote:

---

National Guard Removes Homeless from DC

---

quote:

---

The administration justified the action by citing a public safety emergency and claiming that crime in D.C. was out of control, despite official statistics showing that violent crime had reached a 30-year low and had decreased further in 2025.

---

quote:

---

is crime in dc at a 30 year lowor are they falsifying crime statistics?

---

quote:

---

However, these statistics have been met with skepticism. In July 2025, a D.C. police commander was placed on administrative leave amid allegations from the Fraternal Order of Police (FOP) that crime reports were being downgraded to make crime appear lower.
Union officials allege a broader pattern of manipulation, claiming that officers were directed to classify felony offenses as lesser crimes.
Critics argue this could distort the true picture of public safety in the city.

Despite these allegations, independent analyses using FBI data confirm a significant downward trend in violent crime since 2023.
Homicides in D.C. fell 19% in the first half of 2025 compared to the same period in 2024, a steeper decline than the national average among large cities.
Still, the city's homicide rate remains higher than the average of the 30 large U.S. cities tracked by the Council on Criminal Justice.

In summary, while multiple data sources indicate crime in D.C. is at a multi-decade low, ongoing investigations into potential data manipulation by the MPD have raised questions about the accuracy of the reported figures.

---

quote:

---

The lawyer for a man suing an airline in a routine personal injury suit used ChatGPT to prepare a filing, but the artificial intelligence bot delivered fake cases that the attorney then presented to the court, prompting a judge to weigh sanctions as the legal community grapples with one of the first cases of AI “hallucinations” making it to court.

---

quote:

---

Steven Schwartz—the lawyer who used ChatGPT and represented Mata before the case moved to a court where he isn’t licensed to practice—signed an affidavit admitting he used the AI chatbot but saying he had no intent to deceive the court and didn’t act in bad faith, arguing that he shouldn’t be sanctioned.

---

quote:

---

The use of ChatGPT was discovered after Avianca’s lawyers couldn’t find some of the court documents for the cases to which Mata’s team referred, and filed a request asking for more information on them. Schwartz then went back to ChatGPT to get copies of the case documents, still believing it was a search engine, and handed them over to Mata’s other attorney Peter LoDuca, Schwartz’s attorneys explained in court filings. When reviewing the documents provided, New York-based federal Judge P. Kevin Castel determined the cases were fake, and announced that the court was facing an “unprecedented circumstance” and would consider imposing sanctions.

---

quote:

---

They are definitely not all the same. Gemini and ChatGPT are far ahead of infants like Grok and DeepSeek.

---

quote:

---

ChatGPT =>

Is this true: Gemini and ChatGPT are far ahead of infants like Grok and DeepSeek in AI?

ChatGPT said:
You’re asking whether “Gemini and ChatGPT are far ahead of infants like Grok and DeepSeek in AI.” Let me clarify and update this with the latest benchmarking insights—because terms like “infants” can be misleading here!

1. Who leads where?
Grok 3 (xAI)

Grok 3 achieved breakthrough reasoning and math performance: it was the first model to exceed a 1400 score on Chatbot Arena, outperforming models like Gemini-2.0 and ChatGPT-4o

In benchmarks like AIME 2024, Grok 3 scored about 95.8%, surpassing ChatGPT's advanced models (~93%)

Grok also introduced modes like Think and Big Brain (though the latter isn’t publicly available) to boost reasoning—though some comparison methods (e.g., running the same prompt 64 times and picking a “consensus”) have been critiqued by OpenAI employees

In references tasks, Grok and DeepSeek generated more accurate bibliographic citations than ChatGPT, with none producing fabricated ones—a useful sign of their factual reliability in that domain

ChatGPT (OpenAI)

ChatGPT remains a balanced, versatile model, excelling at creative writing, brainstorming, coding, and fluency across tasks

In visual reasoning tasks (multi-image understanding), ChatGPT-o1 led in overall accuracy (82.5%) and had high consistency (low entropy), placing it above others like Grok 3 in that particular benchmark

Google Gemini

Gemini is a true multimodal powerhouse, handling text, images, audio, video, and even long context—leveraging Google’s ecosystem for real-time research and tools

In specialized fields like dental implantology, Gemini-2.0-flash-Thinking performed best among top models like ChatGPT, DeepSeek, and Grok 3
BioMed Central

Gemini also shows strong math and reasoning benchmarks; for example, on MATH datasets and "Humanity’s Last Exam," it outperformed GPT-4 with ~90% vs mid-70%

DeepSeek R1/V3

DeepSeek uses a Mixture-of-Experts (MoE) architecture, making it cost-efficient and performant in coding and technical reasoning

On general knowledge benchmarks like MMLU, it performs comparably to GPT-4—typically in the mid- to high-80% accuracy range

Its cost-effectiveness has been compared to a “Sputnik moment” in AI—disruptive in both pricing and performance


Answer:Are “Gemini and ChatGPT far ahead” of “infants” like Grok and DeepSeek?

No, that characterization isn’t accurate based on the evidence:

Grok 3 is often ahead in raw reasoning, math, and coding benchmarks—sometimes outperforming both ChatGPT and Gemini.

---

quote:

---

It get answers wrong all of the time. Easy ones too

---