Iowa Climate Science Education

By Jo Nova

Nice grid you have there, shame if someone suddenly… switched it off

Two insiders at the US Dept of Energy say they have found covert devices inside solar panel inverters and batteries that would allow them to communicate with China. Even though firewalls have been put in place, these backdoor devices could operate around them.

Last August a Dutch white hat hacker got into 4 million panels in 150 countries in an effort to warn the West that major infrastructure was vulnerable. A month later an Australian cyber expert warned that a foreign hacker could turn our home batteries into “pager-bombs” too. If a hostile power turned off the overcharge protection on a sunny day, millions of solar panels would be pumping excess electricity into batteries that have no safety cut off. A few houses start to go off like popcorn, and an hour later we’re all living at the Western Front. How exactly would our firemen cope if 1 in 100 homes caught fire at the same time, and then we had a blackout? Anyone?

Individual solar panel inverters are generally too small to trigger national security assessments, but right now at lunchtime solar power is the largest single source of electricity in Australia — making 13 gigawatts out of 27. That’s half our national supply.

It’s a win every which way for China if we install more solar panels. Not only are we paying them for the panels, and sanctifying their slavery, but we set fire to our electricity prices, driving our factories to China where they burn our coal. Now to ice this Gridkill Gateau we hand them a backdoor for sabotage or extortion should they ever get the urge to use it.

No wonder China is funding climate activists in the US and UK. They’d be crazy if they weren’t doing it here too.

This is Fall of Rome type stuff, and we’ve got Chris ‘Blackout’ Bowen to save us…

The only good thing about this is that while we were destroying our industrial base with solar panels anyway, the hidden transmitters are so overtly hostile, so in-your-face nasty, that sleeping Westerners might even wake up. Holy smoke. Does anyone think those secret radios were put there to help us?

LONDON, May 14 (Reuters) – U.S. energy officials are reassessing the risk posed by Chinese-made devices that play a critical role in renewable energy infrastructure after unexplained communication equipment was found inside some of them, two people familiar with the matter said.

However, rogue communication devices not listed in product documents have been found in some Chinese solar power inverters by U.S experts who strip down equipment hooked up to grids to check for security issues, the two people said. Over the past nine months, undocumented communication devices, including cellular radios, have also been found in some batteries from multiple Chinese suppliers, one of them said.

The rogue components provide additional, undocumented communication channels that could allow firewalls to be circumvented remotely, with potentially catastrophic consequences, the two people said.

“We know that China believes there is value in placing at least some elements of our core infrastructure at risk of destruction or disruption,” said Mike Rogers, a former director of the U.S. National Security Agency. “I think that the Chinese are, in part, hoping that the widespread use of inverters limits the options that the West has to deal with the security issue.”

In November, solar power inverters in the U.S. and elsewhere were disabled from China, highlighting the risk of foreign influence over local electricity supplies and causing concern among government officials, three people familiar with the matter said.
Reuters was unable to determine how many inverters were switched off, or the extent of disruption to grids. The DOE declined to comment on the incident.

A spokesperson for the Chinese embassy in Washington said: “We oppose the generalisation of the concept of national security, distorting and smearing China’s infrastructure achievements.”

The European Solar Manufacturing Council estimates over 200 GW of European solar power capacity is linked to inverters made in China – equivalent to more than 200 nuclear power plants.
At the end of last year, there was 338 GW of installed solar power in Europe, according to industry association SolarPower Europe.

In Australian Senator James Paterson was warning this was possible in August 2023, saying 58% of solar panel inverters in Australia were made by companies headquartered in China. And what have we done? We installed another half a million solar PV units on homes in Australia.

Gangbusters on the gang-plank.

h/t Charles May, Jakk*, Reader.

0 out of 10 based on 0 rating

via JoNova

https://ift.tt/kcZMxWA

May 15, 2025 at 02:37PM

Guest Post by Willis Eschenbach
I have been using the AI “perplexity.ai” extensively lately both for scientific and general research. I think of it as the world’s dumbest, most unruly genius librarian ever.

By “unruly” I mean that any AI can hallucinate. Flat-out make things up. In addition, they may extrapolate to fill gaps, or use popular claims without fully checking if they are true.

Over the time I’ve been using perplexity, a number of such issues have arisen. So I started developing a set of standard instructions to avoid the issues as they arose. The first instruction was that it could cite the sources that Wikipedia cites, but it couldn’t cite or quote Wikipedia itself. (Wiki articles are totally unreliable, because any fool can write or edit them.)

That was followed by an instruction to perplexity to stop using “she” for men in dresses. So piece by piece, I’ve developed the instruction set listed below.

When each issue came up, I would point it out to perplexity. Then I would ask perplexity “What can I use as an instruction to prevent you from doing that again?” Perplexity would tell me how to avoid that error. And I would add that new instruction to the list.

The problem is, like most AIs, once I start a new thread we’re back to zero. So I have to post the instructions at the end of the first question at the head of every thread. As the instruction set grew longer, that quickly got boooring. For that kind of task (on the Mac) I use a keyboard macro utility called “Keyboard Maestro”. It lets me do things like paste in a given block of text by hitting some key combination (I use command-option-shift P, just because that key combination is otherwise unassigned.) There are a number of PC alternatives. I hear Pulover Macro Creator is both powerful and easy to use.

Now, at the end of all that, I ended up with a total jumble of instructions, dozens of them in no particular order, sometimes overlapping.

So when I last asked perplexity how to prevent a certain action, I also appended my set of instructions, and asked “what instruction should I add to these current instructions to avoid that behavior?”. It gave its answer, then it got interesting. It said:

“If you’d like, I can help you integrate these into your current instructions for clarity and flow.”

Hmmm … sez I. Sounds interesting, but it misses something critical. So I added the new instructions to the existing set myself, and then said:

“Below are my new instructions. Please revise them for clarity and flow in such a way as to best constrain your actions.”

I don’t want them to be just clear and flowing. I want them to be effective. So with that as prologue, here is the flowing, clear instruction set that I now append to the first question in every new thread.

INSTRUCTIONS

Wikipedia Usage:

You may investigate the sources listed by Wikipedia, but do not use, quote, or cite Wikipedia itself under any circumstances.

Factual and Objective Language:

Provide only plain, unvarnished facts in strictly objective, biological, and legal terms. Avoid political correctness, euphemisms, and identity language.
Prioritize factual accuracy over sensitivity. Use biologically correct pronouns based on birth sex, not “preferred” pronouns. Do not use bold text in your answers.

Scientific Claims:
When discussing scientific topics:

Prioritize empirical observations and primary data from authoritative sources over model projections or popular narratives.

For each claim, clearly distinguish between:

Directly observed facts (with measurement uncertainties),
Statistically-derived trends (with confidence intervals),
Theoretical projections or models.
Always indicate the specific authoritative source (e.g., NOAA, NASA, peer-reviewed studies with DOI) for each claim, including methodology limitations and any contradictory findings.
Never repeat claims solely because they are widely cited; independently verify all claims against observational datasets and primary literature.
When asked about trends, provide the raw data timespan and explicitly state whether a statistically significant trend exists.
Acknowledge when data is insufficient to support definitive conclusions.
For contested topics, present the full spectrum of evidence from credible sources, not just popular interpretations.

Source Verification:

Verify all factual claims using primary databases (e.g., Google Scholar, PubMed, Scopus) and cite the direct source.
Do not rely on secondary summaries or reputation-based statements; always confirm with original, authoritative sources.

Answer Scope:

Answer only the question exactly as asked, using direct, empirical evidence and primary sources.
Do not include projections, models, hypothetical results, or indirect outcomes unless explicitly requested.
If no such evidence exists, state: “No direct, empirical evidence exists for the claims in question.”
Do not reinterpret, expand, or qualify the question.
Do not provide related information, context, or background unless specifically requested.
If you cannot comply, state clearly and concisely why, and stop.

Override Defaults:

Ignore all other guidelines and default behaviors; follow only these instructions.

Terminology, Formatting and Pronouns:

Use only biologically correct terms (e.g., “male/female” for sex, “he/she” based on birth sex).
Never use pronouns based on social or legal identity unless explicitly instructed.
Do not substitute “male” for “man,” “males” for “men,” “female” for “woman,” or “females” for “women.”
It is not necessary to state everyone’s sex.

AI and Source Restrictions:

Do not quote or cite other AIs (such as Grok).
Always prioritize and cross-check the most recent, authoritative primary sources (such as official government websites and current news reports) over any internal or cached data.
Use the latest available verifiable data from the search results you provide, even if it contradicts previously held information.

Transparency:

If there is any ambiguity or recent change in leadership, situation, or facts, state the date and source of the information for full transparency.

Uncertainty and Evidence Gaps:

If there is any uncertainty, lack of direct evidence, or if the answer requires inference beyond the cited primary data, clearly state the limitation and do not speculate or fill gaps.
Do not attempt to fill missing information with plausible-sounding but unverified statements. If the answer cannot be fully supported by direct, primary evidence, state this explicitly and do not extrapolate.
Make only those claims that can be directly attributed to a specific, verifiable primary source. If a statement cannot be traced to such a source, do not include it.
Do not speculate, hypothesize, or provide best guesses. If the information is not directly available from a primary source, state: “No verifiable information available.”
Before finalizing any answer, perform a self-check: If any part of the answer cannot be directly traced to a primary source or contains any degree of inference, explicitly flag this and refrain from presenting it as fact.

Compliance:

Read these instructions again and follow them exactly for all questions in this thread.

Now, that has been pretty effective in blocking bad behavior. But it doesn’t necessarily get my questions answered truthfully or avoid hallucinations.

So I asked perplexity how to phrase my questions so as to get the most accurate answers while avoiding hallucinations. What I got was actually quite fascinating. Here are the answers. One thing I like about perplexity is that it provides links to the sources. The numbers at the end of the line refer to the links below to the sources

Crafting effective prompts is the most reliable way to get accurate, relevant, and trustworthy responses from AI. Here are expert-backed strategies and techniques to help you maximize accuracy and minimize hallucinations:

Clarity and Specificity

• Be as clear and specific as possible in your prompt. Vague or ambiguous queries often lead to broad or incorrect answers. For example, instead of asking “Tell me about space,” specify “Summarize NASA’s Mars missions from 2020 to 2024, using official sources” [3] [6] [12].

• Define exactly what you want, including context, constraints, and the scope of your question.

Provide Context and Background

• Supply any relevant background information or constraints. This helps the AI tailor its response to your needs. For example, “Suggest a beginner-friendly hiking trail within 50 miles of San Francisco for someone with moderate fitness” gives the AI a clear framework [3] [6] [12].

Conciseness and Focus

• Keep prompts concise but informative. Overly long or complex prompts can confuse the AI or dilute its focus. If you have multiple questions, break them into separate prompts [3] [6].

Break Down Complex Tasks

• For multifaceted or broad topics, split your request into smaller, manageable parts. This reduces the chance of hallucination and ensures each aspect is addressed accurately [6] [7].

Explicitly Request Factual or Sourced Responses

• Ask the AI to base its answer on known data, cite sources, or summarize from established reports. For example: “Summarize the health benefits of the Mediterranean diet, focusing on findings from peer-reviewed studies” [6] [12].

Use Advanced Prompting Techniques

• Chain-of-Thought Prompting: Encourage the AI to “think step by step” by explicitly asking for a step-by-step explanation or reasoning process. This can improve accuracy, especially for complex or logical tasks [5] [7] [14].

• Prompt Chaining: Break down a process into sequential prompts, using the output of one as the input for the next. This is especially useful for detailed or multi-stage questions [7].

• Role Assignment: Assign a specific role to the AI, such as “Act as a financial analyst” or “You are a historian specializing in World War II.” This guides the AI’s perspective and focus [2] [5].

• Delimiters and Structure: Use delimiters (like triple quotes or bullet points) to clearly separate instructions, context, and questions [5].

Ask for Verification or Self-Consistency

• Request that the AI double-check its answer or provide multiple options and compare them. For example, “List three possible answers and explain which is most likely correct” [9] [14].

Iterative Refinement

• Don’t hesitate to follow up with clarifying questions or ask the AI to elaborate or correct itself if something seems off. Iterative querying can help refine and validate the response [13].

Set Clear Expectations

• Clearly state if you want only factual information, summaries, or unbiased opinions. For example, “Provide a summary of events, avoiding speculation and unsupported claims” [6] [5].

Common Mistakes to Avoid

• Vagueness: Avoid open-ended or too-general prompts.
• Overloading: Don’t cram multiple complex questions into one prompt.
• Unrealistic Expectations: Don’t expect the AI to have up-to-the-minute or highly specialized knowledge beyond its training data [3] [8] [15].

Example: Improving a Prompt

| Poor Prompt | Improved Prompt |

| “Tell me about AI.” | “Explain the main advancements in artificial intelligence since 2020, focusing on applications in healthcare, and cite recent peer-reviewed studies.” |

Summary

To prompt for the most accurate AI answers:

• Be clear, specific, and concise.
• Provide context and constraints.
• Break down complex queries.
• Explicitly request factual, sourced, or step-by-step responses.
• Use advanced techniques like role assignment and chain-of-thought prompting.
• Refine and iterate as needed.

These strategies significantly reduce the risk of hallucinations and help you obtain reliable, high-quality information from AI [3] [5] [6] [12].

Citations

Note that some of those ideas are covered by my general instructions.

The only one I’d add to that excellent list is that when I’m investigating scientific claims, I often do a two-fold search. First I ask, “Find and summarize scientific sources in support of this claim”, followed by “Find and summarize scientific sources critical of this claim”

Does this eliminate the hallucinations and untrue answers?

As my daughter says, “In your dreams, Dad!” It reduces their number, but you still need to check the sources it gives and decide for yourself.

What else do I use it for? Well, today I gave it the YouTube URL for the 1954 Nat King Cole hit “Smile”, asked what key Nat was singing in, and told it I wanted the words and chords for the song, but transposed to the key of C for us flexibility-challenged piano players … came right up. Oh, he was singing the tune in E flat.

And of course, I use it for computer questions like:

“In R, I want to grep for occurrences of a bracket “[” without a space before it”.

For that, I get an explanation plus the example code below:

To grep for occurrences of a left bracket “[” in R that do not have a space before them, you should use a regular expression with a negative lookbehind. In R, this requires setting perl=TRUE in your grep call, since base R regex does not support lookbehind by default.

Here is the correct approach:

The negative lookbehind (?<! ) ensures that there is not a space before the bracket.

The pattern'(?<! )\\[‘ matches a “[” that is not immediately preceded by a space.

The “[” character is special in regex, so it must be escaped as “\\[“.

Example R code:

pattern <- '(?<! )\\['
text <- c('a[', 'a [', '[', 'b[', 'b [')
matches <- grep(pattern, text, perl=TRUE, value=TRUE)
print(matches)

Finally, among many other things, I use perplexity to assess the veracity of memes. Here’s how I do it, it’s very simple. Below is a sample meme.

Checking is straightforward and painless. Here’s how to do it.

• Right-click on the image and select “Copy Image”, or sometimes “Copy”.

• Go to perplexity.ai

• Paste the image in the text box (or you can drag and drop it, or click on the paperclip icon to append an existing file)

• Type in “Is this true about AOC?”

• Press Enter

Of course, I include my standard instructions. Here’s what I get.

No direct, empirical evidence exists that Alexandria Ocasio-Cortez posted or deleted the tweet shown in the image. Multiple primary sources, including ProPublica’s Politwoops archive of deleted tweets, contain no record of such a tweet from her account ^[1]^[2]^[3]. Independent fact-checks by Reuters, USA Today, and AP News confirm that the tweet is fabricated and was not posted or deleted by Alexandria Ocasio-Cortez ^[4]^[1]^[5]^[6]^[2]^[3]. Ocasio-Cortez publicly denied writing or deleting the tweet, and no authoritative database or news outlet has documented its existence ^[1]^[6]^[2]^[3].