It's easy to get carried away with AI — either by avoiding it completely or by replacing your entire team overnight. But balance is everything. ⚖️
Real signals — like a DM, comment, or social post — can impact your business in ways AI can't always detect. 👀 That's why oversight and human judgment still matter. Trust AI, but always verify. ✅
🎧 Listen to the full episode here 👉 https://youtu.be/sGtPfP7cBao
#AIinBusiness #MarketingStrategy #TrustButVerify #AIBalance...
Sorry, got a bit distracted with Jenkins fun (I do have a day job I think)
This is how we do (since 1982...)
#trustbutverify #sfcon #notjustforrubyanymore #didyouknowyoucanunittestyourchangelogs
People voted Yes *on that basis* to changing our Constitution. Now we are told this solemn commitment can be ditched by the Oireachtas alone. Trust us, they say, while breaking faith with us. #TripleLock #Neutrality #Ireland #Bunreacht #TrustButVerify
AI Search: Yes, absolutely, 100%, go for it 🔥
Reality: Feature not available in your region, browser, planet, or lifetime.
Feels less like 'artificial intelligence' and more like 'confident hallucination as a service' 😌 #AI #UXFail #TrustButVerify
A privacy VPN you can verify
https://vp.net/l/en-US/blog/Don%27t-Trust-Verify
#HackerNews #privacyVPN #verify #onlineSecurity #digitalPrivacy #trustButVerify
Palo Alto: “LOL, we fixed 24 vulns in a random Tuesday update with no CVE alert. Why are you panicking?”
https://security.paloaltonetworks.com/PAN-SA-2025-0012
Just a stealth patch buried in the advisory feed.
The stats:
All bundled into one advisory.
Welcome to the Patch Gacha Machine:
Spin once, fix 24 vulnerabilities (maybe).
PAN CVEs age like wine… and compromise like whiskey.
#PANOS #CyberSecurity #CVEs #PatchAndPray #SilentFixes #PSIRTFail #MemeSec #BlueTeamLife #TrustButVerify
Ah yes, another startup solving the trust crisis in AI by... naming it louder. NeuralTrust's "Echo Chamber" sounds less like a security model and more like what happens when a product team feeds back its own pitch deck for 6 months. 🔁🤖 #AI #TrustButVerify
😬 A cautionary tale for recent grads: A Utah law clerk was fired after using ChatGPT to draft a court filing that included fake legal citations — the first AI-generated “hallucination” officially sanctioned by a Utah court.
🧑⚖️ Judge: Lawyers have a duty to review filings
📉 Law firm: Fired the clerk + added an AI policy
📱 Broader lesson: Overreliance on AI can harm your credibility
🎓 Professors report students increasingly trust AI without verifying
This signals a shift: AI use is no longer optional — but knowing how to check its output is critical.
#AI #LegalTech #ChatGPT #WorkplaceEthics #ProfessionalDevelopment #TrustButVerify
https://arstechnica.com/tech-policy/2025/06/law-clerk-fired-over-chatgpt-use-after-firms-filing-used-ai-hallucinations/
🕯️ Behind the Code: Why Open Source Matters for Small Business Survival #OpenSource #Cybersecurity #SmallBusinessSecurity #DigitalFreedom #TransparentTech #PrivacyMatters #SecureByDesign #FOSS #InfoSec #SecurityAwareness #DataProtection #ZeroTrust #AuditTheCode #OpenSourceSecurity #TrustButVerify #TechTransparency #MatrixProtocol #Passbolt #Bitwarden #Nextcloud #OpenSourceTools #SecurityFirst #DigitalSovereignty #TechEthics #SecurityForAll
Just a few days back we were discussing hbom is tricky coz once we try to read in we are going to tamper warantties and today @bunnie.org Released https://bunnie.org/iris hoping this changes the landscape
#hbom #trustbutverify
Buying a dog from a breeder in NY.
We've been in discussions via phone calls and texts and pictures and videos for 6+ months.
Called the credit union to wire transfer payment (this is a stupid expensive dog, I admit that).
PLEASED as punch that they asked me no less than 3 times if I was SURE I trusted the seller, really wanted to wire the money, had a confirmed transaction, etc.
I'm sure we sounded like the average overseas-magic-puppy-farm victim, and fit the SCAM pattern so well, it was really pleasing to see the CU taking all the precautions to make sure I'm not a victim.
Island.io is not open-source, making it a "trust" fail from the start bar. the history of "a private island" is not exactly trustworthy ;> @Bloomberg @katie_roof @kateclarktweets @island_io #browserwars #privacy #security #opensource #trustbutverify https://www.bloomberg.com/news/articles/2025-02-08/browser-startup-island-valued-at-4-5-billion-in-coatue-led-round
🦩 LIVING ON THE EDGE... literally!
(@rnbwkat made me do it!)
Just your favorite security flamingo casually ziplining across the Panama City skyline at Cielo rooftop bar! Talk about taking "elevated privileges" to a whole new level!
Pro tip: When you're suspended 40 floors up, you REALLY learn to trust your security harness. Just like how you should trust but verify your security controls!
Gotta say, the view of the Pacific is even more spectacular when you're zooming through the air. Though I did have to tuck my feathers in - nobody wants a wind-swept flamingo crash landing at the bar!
Remember friends:
- Sometimes you have to step (or zip!) out of your comfort zone to get the best perspective on things.
- Just make sure your safety checks are in place first!
#SkyHighSasha #PanamaAdventures #TrustButVerify #cybersecurity
Yo! Everyone, listen up for a second.
I know everyone is trying to help out and spread helpful information, but when the information contains an email or phone number, let's verify.
This post https://kolektiva.social/@MikeDunnAuthor/113872667905918726
It has a phone number only for the state of Washington. When I called, they said they were already overrun by out-of-state calls.
This is effectively a DDoS attack. Their website is also down. I'm not sure if this is a deliberate attack to weaken the immigration services in Washington, maybe it's the next place trump is planning (Speculation, worth looking at.).
It's not hard to verify. It should be second nature for all the IT/InfoSec people here.
@henry The BBC article shown here does not mention AI. The post says that the company blamed a human for a copy-and-paste error using an older post.
🌟 Saw a Toot earlier from someone saying he had assumed AI-generated Alt Text is “better than nothing”—but now he’s got an example where it *isn’t*! 🤔 That’s all he shared, so we’re left to draw our own conclusions.
Is useless AI Alt Text still better than nothing? Or just whimsically entertaining? 🌈😂
Sure, bad AI Alt Text can be worse than nothing in certain cases… but do you think blind folks are putting themselves in those situations? 😉
“Trust but verify.” 🔍 Tools like AIRA’s Access AI let you verify AI Alt Text with a human—perfect for when you need a reliable description to make an important decision. 👥💡
Navigating Restaurant Menus as a Blind Person: The Challenges and Opportunities of Language Models
Dining out can be an experience of discovery and joy, but for blind individuals, simply understanding the menu can be a challenge. Recent advancements in technology are helping to bridge this gap, with language models like GPT-4 enabling users like me to take a photo of a restaurant menu and generate a structured response about what’s available. However, this process is not without its challenges. In this article, we'll explore the steps involved in using a language model to analyze a restaurant menu and identify potential pitfalls and ways to mitigate them.
1. Image Preprocessing
When a blind person takes a photo of a menu, the first task is image preprocessing. The app will typically resize, crop, and enhance contrast to maximize text clarity.
**Challenges: ** Blurry images, glare, and poor lighting can severely affect text readability.
**Mitigation: ** To minimize these issues, keep the camera steady and ensure good lighting. If glare persists, reposition the menu or camera to find an angle that minimizes reflections.
2. Object Detection
After preprocessing, the app identifies text regions within the image using object detection algorithms.
**Challenges:** Text might be missed or misidentified if the font is unusual or if it's overlaid on images or poorly contrasted.
**Mitigation:** Focus on clear text regions and, if possible, crop to only show the menu. Multiple photos can help capture different sections if needed.
3. Optical Character Recognition (OCR)
The detected text regions are converted into machine-readable text using OCR, a key step in translating the visual information into text.
**Challenges:** Distorted characters or unusual fonts may result in recognition errors.
**Mitigation:** Choose menus with simple, clear fonts or ask for a digital version. Some settings may improve OCR performance, especially if the text is skewed.
4. Text Normalization
Once the text is recognized, it's cleaned up to improve readability. This process corrects spelling errors, standardizes formats, and removes irrelevant characters.
**Challenges: ** OCR errors can still persist if the menu's structure or formatting is complex.
**Mitigation: ** Cross-checking with accessible menu versions or staff assistance can help ensure key details are not missed.
5. Natural Language Processing (NLP)
The normalized text is then processed using NLP techniques to classify sections like appetizers or desserts and relate menu items to their descriptions.
**Challenges:** Errors can occur due to ambiguity in wording or inaccuracies in text processing.
**Mitigation:** Verify section titles and descriptions to ensure nothing crucial is missing. Asking staff for clarification can fill in any gaps.
6. Response Generation
Finally, the system compiles the information into a coherent summary, providing a helpful overview of the menu's offerings.
**Challenges:** If errors occur during earlier steps, important details might be omitted from the summary.
**Mitigation:** Multiple photos or supplementary descriptions can offer more comprehensive results.
Conclusion
While language models can significantly improve accessibility for blind people when dining out, challenges remain. Recognizing these challenges and adopting effective strategies to mitigate them can create a better dining experience. By using technology thoughtfully and seeking additional assistance as needed, the joy of eating out becomes more accessible and enjoyable.
#Accessibility #AI #AIsoftheBlind #Blind #ComputerVision #Disability #Innovation #LLM #TrustButVerify
Happy (very punctual) #FiveMinuteFriday. Today we're talking about some changes we are making in light of the #XZBackdoor. tldr; the Trusted Contributors program will be doing a bit more #TrustButVerify. Read more about Trusted Contributors at https://www.puppet.com/ecosystem/contribute/trusted-contributors or download the scanner module at https://forge.puppet.com/puppetlabs/xzscanner and check your infrastructure today.
