tiarafawn

See also visidata for an alternative https://www.visidata.org/

Could the real reason for this be more centered around generation and control of new training data?

I suspect the same for the forced high AI usage quotas for developers at MS etc. We've had multiple generations of models trained on all of the code that's available and there are diminishing returns on how much that data can do for training now. Newly published publicly available data is also made up of a significant portion of slop.

The best way to get fresh training data from real human brains might be to have real humans use your first party tools where you control all of the telemetry.

I call this getting slopped in the face

Could also be interesting in case the idea of under water data centers ever returns

https://en.wikipedia.org/wiki/Project_Natick

Or it could help offshore wind farms provide a more stable/predictable output.

if AGI (artificial general intelligence) or superintelligence do in fact one day arrive, it might not seem like much of a leap at all.

That does not seem like a valid conclusion to draw from the observations in the article.

Somewhat related video by "Kurzgesagt - In a Nutshell" "What Happens if the Moon Crashes into Earth" https://youtu.be/lheapd7bgLA

If superintelligence can be achieved, I'm pessimistic about the safe part.

- Sandboxing an intelligence greater than your own seems like an impossible task as the superintelligence could potentially come up with completely novel attack vectors the designers never thought of. Even if the SSI's only interface to the outside world is an air gapped text-based terminal in an underground bunker, it might use advanced psychological manipulation to compromise the people it is interacting with. Also the movie Transcendence comes to mind, where the superintelligence makes some new physics discoveries and ends up doing things that to us are indistinguishable from magic.

- Any kind of evolutionary component in its process of creation or operation would likely give favor to expansionary traits that can be quite dangerous to other species such as humans.

- If it somehow mimics human thought processes but at highly accelerated speeds, I'd expect dangerous ideas to surface. I cannot really imagine a 10k year simulation of humans living on planet earth that does not end in nuclear war or a similar disaster.

This problem/attack is called "confused deputy". It's surprisingly hard to find a link that correctly explains the problem and its mitigations. This one is correct but not very verbose: https://medium.com/@fhbro/confused-deputy-c9e75eb7df00#8edf

Reminds me of the German poem "Storch und Schleiche" [1] (LLM translation):

  A stork once strolled by the pond's edge,
  where he spotted a blind slow-worm, a hedge.
  The stork exclaimed, "Oh, this is quite grand!"
  and devoured it, all, without a reprimand.

  The slow-worm lay in the stork's belly,
  and both found this situation rather smelly.
  The blind slow-worm then spoke with dismay,
  "Such horror!" and slipped through the back door away.

  The stork, vexed by this unhappy twist,
  couldn't believe he'd let the slow-worm persist.
  So, without hesitation or delay,
  he ate the slow-worm again the same way.

  He cleverly wedged the back door tight,
  so the blind slow-worm could not escape his sight.
  The stork called in, with a cunning tone:
  "Well, if you can, try escaping on your own!"

  The sly slow-worm found the front door clear,
  and promptly made his escape with no fear.
  But the stork, filled with anger and spite,
  ate the slow-worm once more, sealing his plight.

  In a crafty invention, the stork's mind did revel,
  he connected both doors to secure his prey well.
  Then, addressing the slow-worm inside without fail,
  "Now get ready for a round trip, without fail!"

They made plans to enforce manufacturers to provide updates for longer (3/5 years for OS/security updates) [1]. Though you might argue that 3/5 years is not enough, this seems like a more effective way to elongate the lifespan of an average user's phone who might not be comfortable with installing and using an alternative OS.

Not sure what the latest update on this is. IIRC the planned penalty was to make the manufacturer give full refunds to the consumer for devices that violate the requirement.

[1] https://www.androidauthority.com/eu-smartphone-updates-rules...

The end game here is developing a mind reading device. The endeavor device is ethically questionable because such a device would have a lot of ethically wrong/questionable applications.

Historical common-crawl data [1] is available for download for free. Their data was the single most impactful source for GPT-3 [2]

[1] https://commoncrawl.org/the-data/get-started/

[2] https://en.wikipedia.org/wiki/GPT-3#Training_and_capabilitie...

I think what the parent was trying to communicate (and what I'm thinking as well) is doubting your premise in 1. ("the model must be thinking beyond the next token").

Rephrase "The model is good at picking the correct article for the word it wants to output next" to "After having picked a specific article, the model is good at picking a follow-up noun that matches the chosen article". Nothing about the second statement seems like an unlikely feat for a model that only predicts one word at a time without any thinking ahead about specific words.

Not good

``` This company helps people create websites by using a tool called GPT-3. They also provide updates and information from their Twitter account @krishnerkar. ```

The only available patch is a major version jump from <= 8.5.1 to 9.0.0. Also "You may need some additional configuration"

See also https://www.howmanydayssinceajwtalgnonevuln.com/

because rent and salaries are paid in something that actually has value: the dollar.

In 2018, 90% of employees received their salaries in BNB

https://www.ccn.com/90-of-employees-at-major-crypto-exchange...

In my Google search just now, the very first result for the search term "aws t2 vs t3" was this site [1]. Google search result rankings vary based on localization, cookies, A/B tests, etc., so it might not be at the very top for everybody.

The content of the website looks much like AI generated nonsense that has not been reviewed by anyone even remotely familiar with the subject matter:

When it comes to thyroid hormones, there is a lot of confusion and debate. Some people swear by the benefits of thyroid hormones T3, while others swear by the benefits of thyroid hormones T2. So, is T3 better than T2 AWS?

[1] https://www.websitebuilderinsider.com/what-is-the-difference...

Parent did not imply that nobody should use them, but maybe offered some valid points to explain why they might not be seeing a very wide adoption. Your response seems more hostile than it needs to be.

I know someone who uses pinterest in a similar way (saving places to see while travelling). Not sure how big the demand for this is in general, but they would also be interested in a location data and a map view for saved items.

Content moderation (which seems to be the biggest point of contention with Twitter right now) is decentralized.

By choosing your hoster, you delegate content moderation capabilities for your feed to them. If you don't agree with their policies, you can switch to a different hoster without having to leave the ecosystem.

I am not sure to which degree this is practiced/convenient in Mastodon currently, but its distributed nature enables this as an elegant solution to the content moderation problem IMO

Twitter’s engineers can no longer make changes to code as of noon in San Francisco, the people said.

I'm quite curious about the

- Structure: Deployment freeze VS full-on halt of development on feature branches

- Enforcement: Mere policy VS technical block via permission revocation (what happens if an urgent hotfix is required?)

- Communication: Did anyone see this coming? Was it communicated in a respectful manner?

Any Twitter engineers here who are able to talk about any of this?

This doc page [1] seems to have all the interesting API methods listed that could be exploitable. Looks like an attacker would need to be able to inject a malicious payload like `${script:xyz}` inside of a String template used for one of the functions that ultimately do String interpolate/replace/lookup actions. While not as trivial as the exploit path for Log4j, it seems conceivable that some applications have injection points here, especially if they perform multi-stage/recursive replace operations.

What's quite interesting is that the `env` Lookup is still enabled by default. If I understand correctly, this would imply that leaking environment variables is still possible even with the fix, if the attacker has an injection point and access to the return value of the vulnerable function.

[1] https://commons.apache.org/proper/commons-text/apidocs/org/a...

Seems exploitable for phishing

You're right, my mistake. Happy to see though that the current Wikipedia page cites an archived version of a first-hand source though rather than FB's suggestion

The original blog post would be a better link to post here rather than the article: https://tech.fb.com/artificial-intelligence/2022/07/how-ai-c...

Reading through it, I strongly disagree with FB's example for "Better citations in action". I don't see an improvement in the wording and IMO they would be making it worse by switching from an official first party source to a third party one.

While their tool might be useful to find semantic (mis)matches, a much more important part of verifying citations is to verify that the source has any business to make claims about the matter in the first place. https://xkcd.com/978/

QUOTE Better citations in action Usually, to develop models like this, the input might be just a sentence or two. We trained our models with complicated statements from Wikipedia, accompanied by full websites that may or may not support the claims. As a result, our models have achieved a leap in performance in terms of detecting the accuracy of citations. For example, our system found a better source for a citation in the Wikipedia article “2017 in Classical Music.” The claim reads:

“The Los Angeles Philharmonic announces the appointment of Simon Woods as its next president and chief executive officer, effective 22 January 2018.”

The current Wikipedia footnote for this statement links to a press release from the Dallas Symphony Association announcing the appointment of its new president and CEO, also effective January 22, 2018. Despite their similarities, our evidence-ranking model deduced that the press release was not relevant to the claim. Our AI indices suggested another possible source, a blog post on the website Violinist.com, which notes,

“On Thursday Los Angeles Philharmonic announced the appointment of Simon Woods as its new Chief Executive Director, effective Jan. 22, 2018.”

The evidence-ranking model then correctly concluded that this was more relevant than Wikipedia’s existing citation for the claim. /QUOTE

Nothing wrong with that... https://csvchain.com/

If they need to rely on their own GPS and internet uplink rather than just bluetooth, they would be much more expensive

I agree that the number of different technologies here is shocking, but you get a lot more value as a result than "just to run your applications". Most notably that includes CI/CD, scaling, and redundancy, all with IaC. These things used to require a lot more administration and development effort.

Plugging together all these different tools has become so much work though that in many organizations the platform team(s) who are mainly occupied with doing just that take up a lot of engineering resources.

In my opinion the next evolutionary step would be for all of this to be bundled and abstracted away. Funnily enough, we pretty much have that product already with autoscaled "serverless" cloud services, GitHub Actions/Azure DevOps pipelines etc.

The biggest problem is probably that things like AWS Lambda, SQS, and DDB lack versatility and user friendliness. If we get some improvements on that front, many organizations might opt for that instead of dealing with their own K8s deployments. Even better would be if we had something like a stripped-down version of OpenStack, just focussed on doing "serverless" right, and rebuilt from the ground up.

By purchasing an NFT you have the right to say that you are the person that purchased that NFT.

Complete and accurate.

You would need Teams installed AND an application that opens the malicious link. IE11 and Edge Legacy do that without prompting the user, other browsers display a confirmation dialog. There is a patch addressing the specific exploit path via MS Teams.

The underlying argument injection in LocalBridge.exe (which is the binary processing the JSON payload) is still present, which can be exploited to open other office apps with injected command line arguments. Someone might find another way to run arbitrary code using command line switches other than --gpu-launcher