Videos btxGmN8RvNU
Your Agent's Biggest Lie: "I Searched the Web" — Rafael Levi, Bright Data
Scene timeline
54 shot(s).
keyframes kept every frame deduplicated
What was stored
- cues
- 241
- whisperx 241
- chunks
- 27
- from 241 cues
- keyframes
- 27
- kept of 54 captured
- frames with text
- 27
- 1,623 lines read
- chapters
- 0
- from the source metadata
- keyframe bytes
- 6.4 MB
- word timings on 241 cues
Provenance
| stage | state | model | started | took |
|---|---|---|---|---|
fetch |
done | — | 2026-08-11 10:30 | 1m 08s |
stt |
done | — | 2026-08-11 10:31 | 17s |
chunk |
done | — | 2026-08-11 10:31 | 0s |
text_embed |
done | — | 2026-08-11 10:31 | 0s |
keyframe |
done | — | 2026-08-11 10:31 | 1m 26s |
ocr |
done | — | 2026-08-11 10:33 | 27s |
frame_embed |
done | — | 2026-08-11 10:33 | 4s |
Frames, and what the machine read
-
- AlEngineer0.98
- EUROPE1.00
-
- PRESENTINGSPONSOR1.00
- Google DeepMind1.00
-
- PLATINUM SPONSORS0.98
- # Braintrust0.96
- WorkOS OpenAI0.95
-
- bright data1.00
- Your Agent's Biggest Lie:1.00
- "I SEARCHED0.99
- THE WEB"0.95
- Rafoel Lavi0.69
- Keep1.00
- licd1.00
- AlEngineel0.92
- EUROPE1.00
-
- x0.50
- docs.google.com/presentation/d/15BjYcbRwfm0tvFbbPY8jt0S5F3Lh8ob/edit?slide=id.g3d4a2d6a79d_0_0#slide=id.g3d4a2d6a79d_0_00.99
- Ask Google1.00
- Claude1.00
- Claude (MCP)0.98
- Claude0.98
- Claude0.90
- Claude1.00
- Claude0.99
- Claude1.00
- Claude1.00
- Claude0.99
- 品0.73
- cy0.73
- All Bokmarks0.94
- Session1_Agents_Biggest_Lie0.97
- File Edit View Insert Format0.99
- Slide1.00
- PPTX1.00
- Arrange1.00
- Tools1.00
- Working offine0.96
- Extensions0.99
- Help1.00
- 回0.54
- Slideshow1.00
- Share1.00
- @ Fit0.80
- 田0.96
- Background1.00
- Layout1.00
- Transition1.00
- (Ctrl-FS)0.91
- bright data1.00
- AlEngineer0.98
- Sign up and get a bonus0.98
- ★0.97
- ★1.00
- AIE1.00
- ★1.00
- ★1.00
- SESSION11.00
- ★1.00
- ★1.00
- Your Agent's Biggest Lie:1.00
- ★1.00
- ★1.00
- ★1.00
- 'ISEARCHED0.95
- A0.98
- THE WEB"1.00
- 60%+1.00
- Rafael Levi1.00
- DevRel EngineerI Bright Data0.99
- LIVE DEMO0.99
- Enhance this slide ×0.96
- 品0.88
- TIMING: 30 seconds1.00
- 日0.68
- Q Search0.99
- Partly sunny1.00
- 19℃1.00
- ENG1.00
- @0.55
- 11:00 AM0.98
- 4/9/20260.97
- Keep1.00
- licc0.89
- Engineering the future of Al0.99
-
- e.com/presentation/d/15BjYcbRwfm0tvFbbPY8jU0S5F3Lh8ob/edit?slide=id.g3d4a2d6a79d_0_239#slide=id.g3d4a2d6a79d_0_2390.98
- AskGoogle0.90
- D0.69
- Claude (MCP)0.99
- Claude1.00
- Claude0.99
- Claude0.98
- 品0.70
- □αy0.55
- Flowers0.95
- ABoolmals0.74
- Session1_Agents_Biggest_Lie0.98
- File Edit View Insert Format1.00
- Slide1.00
- Arrange1.00
- Tools1.00
- Working offine0.98
- Extensions1.00
- Help1.00
- 回0.62
- Slideshow1.00
- Share1.00
- @ Fit0.82
- T00.59
- 田0.93
- Background1.00
- Layout1.00
- Transition1.00
- ★1.00
- AIE1.00
- ★1.00
- ★1.00
- ★1.00
- Al agents get blocked,1.00
- ★1.00
- ★1.00
- ★1.00
- fed fake content, and hit CAPTCHAs.0.99
- Then they report back as if nothing went wrong.0.99
- G0.62
- 60%+1.00
- bright data0.98
- LIVE DEMO0.99
- Enhance this slide ×0.95
- 品0.86
- Click to add speaker notes0.98
- Q Search0.98
- Partly sunny1.00
- 19℃0.99
- ENG1.00
- 11:00 AM0.99
- 4/9/20260.97
- 80.54
- Keep1.00
- licd0.92
- Braintrust0.99
- WorkOS OpenAI0.94
-
- x0.61
- v/presentation/d/158jYcbRwfm0tvFbbPY8jt0S5F3Lh8ob/edit?slide=id.p2#slide=id.p20.93
- Ask Google1.00
- Claude1.00
- Claude (MCP)0.98
- Claude1.00
- Claude0.99
- Claude1.00
- Claude0.99
- Claude0.98
- Claude1.00
- Claude0.92
- Claude0.99
- 品0.62
- αy0.66
- Flowers0.97
- All Bookmarks0.93
- Session1_Agents_Biggest_Lie1.00
- File Edit View Insert Format0.98
- Slide1.00
- PPTX0.99
- Arrange1.00
- Tools1.00
- Extensions1.00
- Help1.00
- 回0.81
- Slideshow1.00
- Share1.00
- C0.54
- 日0.72
- Background1.00
- Layout1.00
- Theme1.00
- Transition1.00
- The Web Is Fighting Your Agent0.99
- And your agent doesn't tell you it lost.0.99
- ★1.00
- ★1.00
- AIE1.00
- ★1.00
- ★1.00
- ★1.00
- Blocked by Default1.00
- Actively Poisoned1.00
- ★1.00
- ★1.00
- ★1.00
- Cloudflare blocks Al crawlers on 20%0.99
- of the web. No config needed by site0.99
- Cloudflare's Al Labyrinth feeds fake1.00
- content to Al agents. Your agent gets1.00
- 圆0.90
- owners. Your agent hits a wall before0.99
- it starts.1.00
- Plausible-looking disinformation.1.00
- data back - it's just not real.0.96
- 690.71
- 口0.82
- 60%+1.00
- 20%1.00
- of the web blocks0.98
- FAKE1.00
- content served1.00
- Al crawlers0.99
- to Al agents0.96
- bright data0.98
- LIVE DEMO0.99
- Enhance this slide X0.97
- 品0.88
- TIMING: 90 seconds1.00
- 日0.89
- Q Search0.98
- Partly sunny1.00
- ENG1.00
- @0.77
- 4/9/20260.97
- 11:01 AM1.00
- Keep1.00
- lic0.95
- AlEngineer0.97
- EUROPE1.00
-
- File Edit Selection View Go Run Terminal1.00
- Help1.00
- workshop1.00
- EXPLORER1.00
- 02_with_mcp.py1.00
- ×0.71
- WORKSHOP0.94
- C0.50
- 02_with_mcp.py >.0.93
- 0 .emv0.69
- 691.00
- env.example1.00
- 700.92
- 800.60
- 02_with_mcp.p0.99
- 01_without_mcp.py1.00
- 711.00
- 721.00
- # IDENTICAL task + systom prompt to Demo 10.97
- 03_live_challenge.p: C:\Users\skurg\Desktop/\Coding\Gt-Hubrightdata-hack-pack(workshop\02_with_mcp.py0.94
- 田0.94
- ① README.md0.96
- results_comparison.md1.00
- requirements.txt0.98
- 751.00
- I need you to complete these 5 tasks. For each one, try your best to get0.99
- the data. If you can't, explain what went wrong.0.99
- results_without_mcp.md1.00
- 79 8 81 820.75
- 781.00
- PROPERTY - Get rental listings near the conference venue in London from Rightmove:1.00
- List at least 5 specific properties with prices.0.99
- https://www.rightmove.co.uk/property-to-rent/find.htm1?searchLocation=Sw1%2C+South+West+London&useLocationIdentifier=10.99
- ★1.00
- COMPANY - Get the LinkedIn company profile for Mistral AI:1.00
- AIE1.00
- ★1.00
- 831.00
- https://ww.linkedin.com/company/mistral-ai/0.99
- ★1.00
- 3. PROFILES - Get the Instagram profile for @ylecun:0.99
- https://ww.instagram.com/ylecun/0.98
- ★1.00
- ★1.00
- ★1.00
- 4.1.00
- PRODUCT - Get the Amazon product page for this book:0.98
- https://ww.amazon.com/dp/10981079691.00
- 5. SOCIAL - Get the TikTok profile for @google:0.97
- 921.00
- https://ww.tiktok.com/@google1.00
- 931.00
- PROBLEMS1.00
- OUTPUT1.00
- DEBUG CONSOLE1.00
- TERMINAL1.00
- PORTS1.00
- Filter (eg. text, lexclude Text, t.0.96
- Code1.00
- ① A git repository was found in the parent folders of the0.96
- ×0.84
- workspace or the open file(s). Would you like to open the0.99
- repository?0.98
- > OUTUINE0.89
- Source: Git0.98
- Always1.00
- Never1.00
- > TIMEUINE0.94
- 0△。0.60
- Ln 74, Col 11 Spaces: 4 UTF-8 CRLF0.99
- () Python0.98
- 80.55
- Python 3.13.41.00
- Continue (NE)1.00
- C0.64
- Search1.00
- Partly sunny1.00
- ENG1.00
- 。0.60
- 11:02 AM0.99
- 4/9/20260.95
- KeP0.88
- C0.85
- AlEngineer0.97
- EUROPE1.00
-
- File Edit Selection View Go Run Terminal0.99
- Help1.00
- workshop1.00
- 00.59
- EXPLORER1.00
- 01_without_mcp.py×0.99
- 日0.79
- WORKSHOP1.00
- 800.59
- Q1.00
- 0 env0.88
- env.example0.99
- 01_without_mcp.py1.00
- 441.00
- 431.00
- 2.1.00
- COMPANY - Get the LinkedIn company profile for Mistral AI:0.99
- https://www.linkedin.com/company/mistral-ai/0.99
- 02_with_mcp.py0.99
- 461.00
- ① README.md0.96
- 03_live_challenge.py0.97
- 471.00
- 481.00
- PROFILES - Get the Instagram profile for @ylecun:0.98
- https://www.instagram.com/ylecun/0.98
- 田0.90
- results_comparison.md1.00
- requirements.txt0.93
- PROOUCT - Get the Amazon product page for this book:0.98
- https://ww.amazon.com/dp/10981079690.98
- results_without_mcp.md0.97
- 5. SOCIAL - Get the TikTok profile for @google:0.96
- https://www.tiktok.com/@google0.99
- ★1.00
- After all 5 tasks, give a final scorecard: how many succeeded vs failed,0.99
- AIE1.00
- ★1.00
- keep in mind if you found data on a different site it does not count as success.0.99
- 581.00
- ★1.00
- 591.00
- ★1.00
- ★1.00
- ★1.00
- PROBLEMS1.00
- 600.99
- SYSTEM PROMPT = (0.99
- OUTPUT1.00
- DEBUG CONSOLE0.99
- TERMINAL1.00
- PORTS1.00
- Filter (e g. text, 10.91
- Code1.00
- =-|0×0.52
- AGENT RESPONSE (24.7s)0.98
- - Failed: I don't have live web access in this0.98
- 1) PROPERTY - Rightmove (London SW1, rentals)0.97
- chat, and Rightmove's listings are rendered behind JavaScript/CDN protections (Akamai)0.99
- that require a real browser session. Without0.99
- browsing tool or page HTML, I can't extract at least 5 specific properties with prices.0.99
- 2) COMPANY - LinkedIn company profile for Mistral AI0.98
- - Failed: LinkedIn company pages are login-gated and protected by anti-bot measures; without an authenticated browser tool I can't1.00
- load or parse https://www.linkedin.com/company/mistral-ai/. If you can enable browsing or share the page HTML, I'll extract the1.00
- profile details.1.00
- 3) PROFILES - Instagram profile for @ylecun0.99
- - Failed: Instagram profile pages require login and heavy client-side rendering; without a browser tool I can't access or parse1.00
- https://ww.instagram.com/ylecun/. Providing a screenshot or HTML dump would let me extract bio, followers, and recent posts.0.99
- > OUTLINE0.89
- > TIMELINE0.98
- - Failed: Amazon pages are protected by bot checks and dynamic rendering; without live web access I can't retrieve https://www.amazon.0.98
- 4) PRODUCT - Amazon product page (ASIN: 1098107969)0.99
- ②0△。0.52
- Update is ready, click to restart.0.99
- Ln 40, Col 20 Spaces: 4 UTF-8 CRLF () Python Python 3.13.40.98
- Continue (NE)0.96
- C0.86
- Q Search0.98
- Partly sunny1.00
- ENG1.00
- 11:03 AM0.97
- 4/9/20260.98
- Ket0.71
- Engineering the future of Al1.00
-
- File Edit Selection View Go Run Terminal1.00
- Help1.00
- workshop0.96
- 00.87
- EXPLORER0.99
- 02_with_mcp.py ×0.93
- WORKSHOP1.00
- 02_with_mcp.py>0.97
- Q0.99
- 0 env0.84
- env.example0.99
- 791.00
- https://www.rightmove.co.uk/property-to-rent/find.htm1?searchLocation=Sw1%2C+0.99
- List at least 5 specific properties with prices.1.00
- seLocationIdentifier=10.96
- 800.63
- ① README.md0.96
- 02_with_mcp.py1.00
- 01_without_mcp.py1.00
- 03_live_challenge.py0.99
- 831.00
- 841.00
- 851.00
- 3. PROFILES - Get the Instagram profile for @ylecun:1.00
- 2.1.00
- cOMPANY - Get the LinkedIn company profile for Mistral AI:0.99
- https://www.linkedin.com/company/mistral-ai/0.99
- 田0.83
- results_comparison.md1.00
- requirements.txt0.96
- PROBLEMS1.00
- R60.89
- httns://ww.instaaram.com/vlacun/0.91
- OUTPUT DEBUG CONSOLE TERMINAL PORTS0.99
- Code1.00
- C×0.83
- results_without_mcp.md1.00
- Prevlew: [i"id":"mistralal","name":"Mistral Al","countrycode":"rK","1ocatlons":["Parls, trance /5vle, rK"」,"rollowers":62zsy1,0.87
- "employeesinlinkedin":970,"about": "Frontier AI. In your hands. We believe in a futu...0.98
- *★★0.52
- ★1.00
- AIE1.00
- ★1.00
- BRIGHT DATA MCP: web_data_instagram_profiles0.99
- Args: {"ur1": "https://ww.instagran.com/ylecun/"}0.96
- ★1.00
- Result: SUCCESS 3,268 chars0.99
- Preview: [("account":"ylecun","fbid":"17841401555783297","id":"851059859","followers":2716,"postscount":3,"isbusinessaccount":false,1.00
- 2.6s1.00
- ★1.00
- ★1.00
- ★1.00
- "isprofessionalaccount":false,"isverified":false,"avgengagement":0.0384,"exte...1.00
- BRIGHT DATA MCP: web_data_amazon_product1.00
- Args: {"ur1": "https://ww.amazon.com/dp/1098107969"}0.98
- Result: SUCCESS | 12,971 chars | 8.5s0.98
- "sellername":"Amazon.com","brand":"Chip Huyen","description":"Machine learning systems are both c...0.99
- BRIGHT DATA MCP: web_data_tiktok_profiles1.00
- Args: {"ur1": "https://ww.tiktok.com/@google"}0.97
- Result: SUCCESS | 28,912 chars | 17.9s0.98
- Preview: [("accountid":"google","nickname": "Google","biography":"Here to help","awgengagementrate":0.0e29e37617554858934,0.98
- "commentengagementrate":0.00004966562173458725,"1ikeengagementrate":0.002854096133751306...0.99
- > OUTUINE0.88
- > TIMELINE0.95
- Step 2: Calling GPT...1.00
- y0.67
- ②0△。0.56
- Update is ready, click to restart.1.00
- Ln 89, Col 38 Spaces: 4 UTF-8 CRLF () Python0.98
- Python 3.13.41.00
- Continue (NE)0.98
- Q Search0.98
- 19℃0.91
- Partly sunny0.96
- ENG1.00
- 11:05 AM1.00
- 4/9/20260.98
- Kep0.89
- lc'0.64
- Engineering the future of Al1.00
-
- x0.60
- docs.google.com/presentation/d/158jYcbRwfm0tvFbbPY8jt0S5F3Lh8ob/edit?slide=id.p5#slide=id.p50.98
- Ask Google1.00
- D0.64
- Claude1.00
- Claude (MCP)0.99
- Claude1.00
- Claude0.95
- Claude1.00
- Claude1.00
- Claude0.99
- Claude1.00
- Claude1.00
- Claude0.99
- 品0.62
- Flowers0.98
- New tab0.99
- Ctd+T0.59
- Session1_Agents_Biggest_Lie1.00
- File Edit View Insert Format0.95
- Slide1.00
- PPTX1.00
- Arrange1.00
- Tools1.00
- Extensions Help1.00
- 回0.75
- New window1.00
- New Incognito window1.00
- OH+Sat-N0.56
- Cht+N0.69
- C0.58
- Fit1.00
- Background1.00
- Layout1.00
- Theme1.00
- Transition1.00
- Rafael (Ratael bral)0.85
- Signed in0.97
- Passwords and autofill0.92
- History1.00
- Live Demo0.98
- Downloads1.00
- Cbl+10.84
- Bookmarks and lists0.99
- Same agent. Same sites. Two very different outcomes.0.98
- Tab groups0.97
- Extensions1.00
- ★1.00
- Cbl+Shit+Del0.89
- AIE1.00
- ★1.00
- Zoom1.00
- 100%1.00
- ::0.81
- ★1.00
- ★1.00
- Without Web Infrastructure1.00
- With Bright Data Web MCP1.00
- e0.54
- Print..0.91
- Orl+P0.73
- ★1.00
- ★1.00
- ★1.00
- Search this tab with Google Lens0.98
- · Raw agent HTTP requests0.96
- • Same agent, same code0.97
- B0.58
- Find and edit0.92
- Translate..0.94
- Hits Cloudflare walls0.99
- · Anti-bot bypass handled0.96
- Casit, save, and share0.93
- · JavaScript doesn't render0.96
- • Full JS rendering0.96
- More tools0.99
- 60%+1.00
- • CAPTCHAs block silently0.96
- · CAPTCHAs solved automatically0.97
- Settings0.99
- Help0.94
- · Returns fabricated data0.97
- Returns real, verified data0.98
- Exlt0.77
- bright data0.98
- LIVE DEMO0.99
- Enhance this slide X0.96
- 品0.87
- TIMING: 30 seconds1.00
- 日0.69
- Q Search0.99
- Partly sunny0.96
- ENG1.00
- 4/9/20260.96
- 11:07 AM1.00
- Ke-p0.80
- MC0.72
- Engineering the future of Al0.98
-
- All Bookmarls0.93
- You've gone Incognito1.00
- Others who use this device won't see your activity, so you can browse more privately. This won't0.98
- AIE1.00
- Downloads, bokmarks and reading list tems wil be saved. Learn more0.89
- change how data is collected by websites you visit and the services they use, including Google.0.99
- ★1.00
- Chrome won't save:1.00
- • Your browsing history0.96
- Your activitby might still be visible o:to:0.96
- • Websites you visit0.94
- ★1.00
- ★1.00
- ★1.00
- Cookies and site data0.99
- Information entered in forms0.98
- Your employer or school0.98
- • Your internet service provider0.95
- Third-party cookies are blocked1.00
- When you're in Incognito mode, sites can't use third-party cookies. If a site that relies on1.00
- these cookies isn't working, you can try giving that site temporary access to third-party0.97
- cookies.0.96
- Q Search0.98
- Partly sunny0.96
- ENG1.00
- 11:07 AM1.00
- 4/9/20261.00
- leep0.88
- ic0.72
- Engineering the future of Al0.97
The page's on-screen-text budget of 600
lines is spent, so the last cards in this grid list fewer lines than they
hold. Narrow the page with ?frames= to read them.
Transcript
241 cues· 2,511 words· 13,077 chars
- 0:15 OK, so let's just work with it like this, a small room.
- 0:19 Hi, everybody.
- 0:19 Welcome.
- 0:20 My name is Rafael.
- 0:22 I represent Bright Data.
- 0:24 Bright Data is basically a web access platform to help agents or anybody collect public data on scale.
- 0:31 And I'm here to talk about the LLMs, misleading people, and all the time convincing them, hey, I did a search.
- 0:37 Hey, I did this.
- 0:39 while it didn't.
- 0:40 Why?
- 0:40 Because LLMs are programmed to please people, please users, so they're making up things.
- 0:44 And this is the biggest issue right now I'm seeing with LLMs.
- 0:48 I'm building applications all the time.
- 0:50 I would rather LLM tell me, no, I can't.
- 0:52 But it never does.
- 0:53 It always tries to make things up.
- 0:55 So currently, the web is actually fighting the robots and automations.
- 0:59 And it's been something that's going on for years.
- 1:02 Everybody knows captures.
- 1:04 The first captures showed up a decade ago.
- 1:08 And it just keeps growing.
- 1:10 And now we have AI blocking AI.
- 1:12 And there's a whole world going on.
- 1:13 And the web access is actually not as simple as it looks.
- 1:17 So they're getting CAPTCHAs, and they don't actually report CAPTCHAs.
- 1:22 So it tries to go and find the data a different way.
- 1:25 Sometimes it goes into training data, and this is where the worst thing is, when it uses training data and tells you that this is the current situation.
- 1:34 Training data is from 2024, we're in 2026, and it doesn't add up, right?
- 1:41 These are some of the new things that I was just literally checking out.
- 1:44 So Cloudflare blocks AI crawling for about 20% of the web.
- 1:47 So 20% of the web is literally not accessible by AI by default fetch that's built into it.
- 1:53 And Cloudflare also right now released an AI labyrinth to actually trap bots and mislead them and provide to them fake data.
- 2:01 So then your results are getting even worse.
- 2:07 So the invisible failure route, right?
- 2:08 There's no error, no warning, just wrong answer, right?
- 2:11 So the agent sends the request.
- 2:13 It gets a capture.
- 2:15 Even an empty page, it doesn't tell you, hey, I got an empty page.
- 2:19 It will try to make up something.
- 2:20 And this is where the most of hallucinations come from, the need to please and lack of data.
- 2:26 So it literally makes things up.
- 2:27 I've seen it literally make numbers up, provide fake citations.
- 2:30 You click on the citation.
- 2:31 It's a 404.
- 2:32 The page doesn't exist.
- 2:33 And I'm sure all of you have seen
- 2:35 I've seen that happening recently.
- 2:38 I mean, literally, 60% of citations on ChatGPT is not working.
- 2:43 How many of you have tried to purchase a product?
- 2:45 Hey, find me this product online.
loading