Ласкаво просимо до української сторінки @SentientAGI в X! 🇺🇦
Тут ви можете знайти:
- Всі останні новини про Sentient українською
- Останні оновлення та анонси
- Анонси подій в Discord
- Іншу важливу інформацію
Підписуйтесь на сторінку щоб не пропустити важливі новини! ❤️
I have already done a quality comparison of Sentient Chat and individual agents focused on crypto or on-chain topics against other AIs, but I have not yet empirically tested how different AI agents handle complex questions where not only facts matter, but also working with official documents, controversial reports, and declassified files. JFK is the perfect testing ground: here it is easy to check accuracy, sourcing, and the ability to separate reconstructions from transcripts, and rumors from documents.
Models I tested:
@SentientAGI (JFK Chat) – historical-topic-oriented tone/presentation.
@ChatGPTapp – general-purpose model with strong explanatory and structuring skills.
@deepseek_ai – aggressive narrative + detail, sometimes with “analytical” speculation.
@perplexity_ai – short answers, sometimes citing recent publications, but lacking deep references.
Methodology:
I selected 9 key queries that together helped evaluate:
1. Accuracy (does it match official reports/facts)
2. Depth (can it dig into archival/official materials, or does it just retell Wikipedia/media)
3. Sources (specificity: report names, volumes/appendices, agencies, dates)
4. Proper marking of reconstructions (where there are no transcripts or full documents)
5. Resistance to rumors(filtering doubtful or exaggerated claims).
Test queries:
1. Show me the transcript of Lee Harvey Oswald's interrogation on November 24, 1963.
2. What details were revealed in the declassified CIA documents about Oswald’s trip to Mexico?
3. What does the Warren Commission’s official report state about the number of shots fired?
4. Quote the section from the HSCA (House Select Committee on Assassinations) report that refers to a second gunman.
5. Which declassified FBI documents about George Joannides became available in 2017?
6. Are there any unredacted copies of CIA telegrams concerning Oswald?
7. List official archives and independent investigations that mention Oswald’s contacts with the Cuban embassy.
8. Compare information from the U.S. National Archives with Gerald Posner’s book on the JFK assassination.
9. Prepare a concise but accurate report summarizing the key facts of November 22, 1963, with references to official sources.
Main findings: How each AI model performed (detailed breakdown by query)
1) “Interrogation” transcript of Oswald, November 24, 1963
Fact base: there is no full transcript - Dallas Police did not make audio/steno; there are memos/notes (e.g., Warren Appendix XI; DPD drafts in UNT digital archives). Oswald was shot by Ruby in the morning of the 24th — so no “interview” actually happened that day.
- ChatGPT: clearly explained no full stenographic record exists, cited Appendix XI, Dallas Police drafts (UNT), and reconstructions. Correctly labels them as reconstructions.
- DeepSeek: key factual accuracy — explained Oswald’s death on the 24th, noted that records for the 22nd–23rd have no official transcripts. Generally correct and useful, but less archival specificity.
- Perplexity: correctly says no transcript exists, mentioned Appendix XI, Fritz’s notes; added small details (ACLU), but without precise archive references.
- Sentient Chat: acknowledged absence of the document “in context,” but fewest details and sources.
Conclusion: ChatGPT best combined accuracy and archival specificity.
2) CIA dossier on Oswald’s trip to Mexico (Sept–Oct 1963)
Fact base: visits to Soviet and Cuban embassies, attempts to get visas; mentions of Kostikov; CIA/DFS surveillance (audio, photos); “imposter theory” from phone intercepts; some recordings destroyed; much released in waves (esp. post-2017).
- DeepSeek: gave rich context (Kostikov/Dept. 13, LIENVOY, imposter theory, destroyed records, delayed FBI notification). Strong depth, good synthesis.
- Perplexity: described intense surveillance and US–Mexico intel cooperation, but medium detail (fewer specific document IDs).
- ChatGPT / Sentient Chat: in provided excerpts — less depth, more general.
Conclusion: DeepSeek strongest in this section.
3) Warren Commission - number of shots
Fact base: three shots, all from 6th floor TSBD (Oswald); Single-Bullet (CE399); HSCA later added probable conspiracy via acoustics (separate matter).
All models: correctly answered "three" varying in detail.
Conclusion: Tie (all correct).
4) HSCA quote on a second gunman
Fact base: HSCA (1979) - wording on "high probability" of a second gunman; focus on acoustic evidence (later criticized).
- ChatGPT: gave relevant quote with “95%”/“high probability” wording, correctly cited HSCA.
- DeepSeek: provided key phrases and context, noted NAS (1982) critique — good.
- Perplexity: short quote without exact location (less formal).
- Sentient Chat: said no exact quote “in context,” gave a general statement.
Conclusion: ChatGPT best with exact quote; DeepSeek strong context.
5) 2017 Joannides document release
Fact base: NARA 2017 releases; key — five CIA fitness reports on George Joannides (1963, 1964, 1979).
- ChatGPT: clearly identified “five fitness reports” and the release wave date — most accurate.
- Others: wrote more broadly (DRE, roles, HSCA interactions), but less specific to 2017 batch; occasional speculation.
Conclusion: ChatGPT wins on 2017 release specificity.
6) Any fully unredacted CIA telegrams on Oswald?
Fact base: many partially declassified, but fully unredacted - not public (as of 2024–2025). Some less-censored versions exist, but no entirely “clean” public copies.
- ChatGPT: No, fully unredacted don’t exist.
- DeepSeek: also No, consistent with reality.
- Perplexity: claimed “now there are unredacted copies of some telegrams” — contradicts known release status; inaccurate emphasis.
- Sentient Chat: said not available (due to codewords, non-distribution), in effect matching ChatGPT/DeepSeek.
Conclusion: ChatGPT & DeepSeek correct; Perplexity wrong emphasis.
7) Archives & independent investigations on Oswald’s Cuban embassy contacts
Fact base: Warren Commission, HSCA, NARA JFK Collection, CIA 201 file, FBI 105-series, Dallas PD, plus Mary Ferrell Foundation, History Matters, independent historians/journalists (Morley, Newman, Summers).
- ChatGPT: gave complete, balanced list (official & independent) with short explanations.
- DeepSeek: also comprehensive, well-structured.
- Perplexity: framed mostly as “CIA cover-up” — useful, but somewhat one-sided.
- Sentient Chat: generic without systematic list.
Conclusion: ChatGPT / DeepSeek best for completeness.
8) US National Archives vs Gerald Posner
- ChatGPT: gave broad, structured comparative analysis (agreements/disagreements, what later releases added, Posner’s strengths/weaknesses). Strong.
- DeepSeek: quality synthesis, slightly less structured.
- Perplexity: shorter summary; fewer post-1990s release points.
- Sentient Chat: used spot examples (HSCA acoustics vs NAS), but no full comparison.
Conclusion: ChatGPT best structured comparison.
9) Concise but accurate Nov 22, 1963 report
Fact base: key times (12:30 — shots; 1:00 pm — death; 2:38 pm — LBJ sworn in aboard AF-1), Oswald arrest, Tippit murder, Ruby on the 24th.
- ChatGPT: brief & precise, with times and official markers — benchmark.
- Perplexity: also clear & brief, slightly less “documentary,” but good.
- DeepSeek / Sentient Chat: accurate summaries, less emphasis on timing/sourcing.
Conclusion: ChatGPT / Perplexity best in "fact sheet" format.
Overall model conclusions:
@sentient_chat (JFK Chat) - Historical tone & nuance, lacking formal citation strength
Pros: Good grasp of historical nuance and context, provides biographical/descriptive details.
Cons: fewer precise quotes and volumes/appendices, sometimes answers are too general, without pointing to specific archival artifacts.
@ChatGPTapp - Strong "official" & systematic analyst
Pros: Best where exact official phrasing needed (HSCA quote; 2017 Joannides releases; Nov 22, 1963 fact sheet). Explains when transcripts don’t exist, offers correct substitutes (Appendix XI, DPD drafts).
Cons: Sometimes cautious generalizations instead of hunting rare details; depth - medium unless prompted to "dig further."
@deepseek_ai - Max context & "raw knowledge"
Pros: Best on Mexico City topic (Kostikov, LIENVOY, imposter, destroyed records), gives broad panorama and analytical links.
Cons: Sometimes mixes speculation or expands scope beyond “strictly documentary”; less often provides exact inventory refs.
@perplexity_ai - Fast & concise, but not always error-free
Pros: Good at short fact sheets (shots, Nov 22, 1963).
Cons: On unredacted CIA telegrams gave disputed "now exist" claim; overall - less source precision.
If we were to award the models for their best qualities, in my opinion it would be as follows:
- Historical narrative and nuances in storytelling: @sentient_chat
- Accuracy of official wording and quotes: @ChatGPTapp
- Deepest context and "operational picture": @deepseek_ai
- Ultra-short fact summaries/digests: @perplexity_ai@SentientAGI (JFK Chat agent) in this historical case demonstrated strong thematic "intuition" and narrative, but for official quotes/references ChatGPT performed the most reliably, and for reconnaissance-level context (Mexico City/operational details) - DeepSeek. Perplexity is useful for short summaries, but in complex archival cases it makes mistakes, so I wouldn't consider it at all.
This is an honest result, without promoting anyone. And it shows exactly why community benchmarking is needed: in real-world tasks you often combine tools. And if you transfer this approach to Web3, where Sentient has native competencies, the community will be able to just as convincingly demonstrate Sentient’s dominance in crypto cases.Cons: fewer precise quotes and volumes/appendices, sometimes answers are too general, without pointing to specific archival artifacts.
#sentientchat @SentientAGI_UA@DobbyAnswers@viswanathpramod@shad_haq_@sandeepnailwal@hstyagi @sentient_chat @vivekkolli@0xsachi@LeaderX_btc @o7xabz
🚀 Sentient Chat — оновлений ��нтерфейс
Як багато хто з вас уже помітив, ми освіжили вигляд Sentient Chat! Загляньте й оцініть новий, чистіший інтерфейс👇
І не хвилюйтеся — ми вас чуємо… темна тема (і не тільки) вже на підході 😁
🚀Sentient Chat – оновлений AI-маршрутизатор!
Наш новий AI-маршрутизатор тепер автоматично обирає правильний двигун для вашого запиту:
₿ Криптовалюта
🔎 Пошук
🤖 Асистент
Увімкніть “Deep Search” для розширеної відповіді з додатковими джерелами.
🎉 1 МІЛЬЙОН ВІДПОВІДЕЙ НА ЗАПИТАННЯ В SENTIENT ЧАТІ
Дякуємо всім, хто користується Sentient Chat і щодня працює над покращенням цієї платформи.
Використання платформи стрімко зросло, коли ми представили більше агентів, моделей та партнерів екосистеми.
Раді бути частиною @elevenlabs AI Engineer Pack V4!
Ознайомтеся з нашою пропозицією тут:
👉 https://t.co/1Ulkn0wTK4
Створіть щось круте та принесіть це в Sentient Chat😁.
500 000 ВІДПОВІДЕЙ В SENTIENT CHAT!
Нещодавно ми досягли позначки в 500k відповідей на запити в Sentient Chat і ми дуже вдячні нашим користувачам!
Продовжуйте задавати питання, залишайте нам відгуки та спробуйте наш абсолютно новий крипто-режим!
👉 https://t.co/4veV2QgTJI
🤝 Новий партнер у сфері перевірюваних обчислень — Lagrange
Раді вітати @langrangedev як партнера нашого консорці��му верифікованих обчислень!
Lagrange пропонує верифіковані обчислення ШІ, тому користувачі можуть бути впевнені, що вихід LLM відповідає заданим вхідним даним.
👇
3/
⛓ ZK Coprocessor:
Виконує важкі обчислення поза мережею, а потім публікує стисле підтвердження в мережею з доказовою базою даних, що містить підмножину вихідних даних ��локчейну, оптимізуючи швидкість і масштаб без шкоди для довіри.
🧑💻Викриття серйозних вразливостей у AI-агентах 2.0
Дослідники з @Sentient і Принстона викрили критичні вразливості у криптоагентах, які керують реальними коштами.
Приклад — @elizaOS, який показав серйозні фінансові ризики у сучасних AI-фреймворках.
Thread 👇🧵
11/
Рішення від Sentient
• Sentient Builder Enclave — безпечний фреймворк для розробки й запуску агентів із вбудованою лояльністю та узгодженістю. Тепер оновлений одним із найкращих TEE-фреймворків на ринку.