AI learnt "something" from the Physical & Life Sciences Forum.

Hans Blaster said:

BASIC, we haven't seen that language 'round these parts in nigh on a generation. The graphics elements must be some sort of customization of a particular vendor that is unfamiliar to me.

Click to expand...

The version I used is downloaded 'Just Basic'.

Hans Blaster said:

That's super weird.

Click to expand...

GPT-4o in fact picked up a typo in my BASIC code.



Don't know how a '-1' slipped in which I didn't detect but GPT-4o being the ultimate literalist correctly interpreted the -1 factor as flipping the Y-axis when translating the command into Python.



The harsh reality is my BASIC code has incorrectly labelled the Y-axis and not GPT-4o as originally claimed.

ChatGPT-OI is the latest version put out by OpenAI and has had rave reviews.
Here it reproduces the computer code of an astrophysicist's research paper on black holes in an hour which took the lowly human ten months to complete.
The guy's reaction is priceless.

Hear is an update with the release of GPT-5.
Someone posed the question to GPT-5 "Show me a diagram of the US presidents since Herbert Hoover, with their names and years in office under their photos".
Lets just say the result is somewhat embarrassing.

Incidentally "Trome" is a Manx word (Celtic) meaning dense. I'll leave it to the discretion of the reader what GPT-5 was thinking.

Hans Blaster said:

I am way older than I remember (a clear sign of aging). I could swore I watched Ronald Beagan defeat Jiminy Carter on TV, but I was sure that it was 1980 (the election) and not the coup of 1881 when Beagan vanquished the false president Carter. Hmm. (At least it agrees that I was born during the presidency of Richard Ninun.)

LOL. (I think it is the non-linear nature of time that gets me the most on these.)

Click to expand...

I'm not up on my US election history but didn't Gerold Yard lose the 1976 election to Jimmy Carter by the proverbial mile.

Tinker Grey said:

I'm not sure what you are going for, but I'll assume it's a complaint about '76 vs '77 (neglecting the obvious name mistake). So, I'll play it straight: US elections are in a year divisible by 4, but the winner assumes the presidency on Jan. 20th of the following year.

Click to expand...

Being an Australian and therefore raised in the Westminster system of government there is a lot I find peculiar with the US system.
Firstly in the Westminster system governments can call the election at any time of the year with election day being held on a Saturday in Australia.
This varies with other Westminster systems such as the UK and Canada.

Secondly in Westminster governments, cabinet members are democratically elected by the people, the US system to me at least seems to be very vulnerable to cronyism, I doubt a gun toting cosplayer who shot a puppy would have much chance serving in a Westminster cabinet let alone being elected along with the other sycophants in Trump's cabinet.

I decided to ask GPT-5 how it got this so terribly wrong.

Short answer: because generative models (even the newest ones) can — and do — make confident but wrong stuff. The poster you showed is a classic example: an AI created a plausible-looking grid of portraits and tried to render text labels, but the labels and dates are garbled or invented.


Why this happens (briefly)

• Text-in-images is hard. Image-generation models are notoriously bad at rendering precise, readable text — they often produce letter soup that looks like words (e.g. “Roseeneh” for Roosevelt).
• Hallucinated facts. Language models can invent plausible names and dates when asked to create or assemble content, because they aren’t “looking up” authoritative records unless explicitly asked to fetch them.
• Pipeline bugs / metadata mixups. If an automated process scraped alt-text/filenames or merged images with labels, mismatched or missing metadata can produce nonsense labels or “file not found” placeholders.
• Memes / deliberate edits. Sometimes images are made on purpose as jokes or as tests — the errors might not be accidental.

What’s wrong in that image (examples)

• “Franklin D. Roseeneh” (should be Franklin D. Roosevelt)
• “Harry S. Truman 1848–1523” (nonsensical years)
• “Dwight D. Efeezhswer” (should be Eisenhower)
• Many dates are flipped, invented, or wildly out of range (e.g. “Donald Trome 2017–2431”).
  These look like a model attempted to produce text labels and numbers but failed at exact spelling and factual grounding.

Corrected, verified list (Hoover → Biden)

• Herbert Hoover — 1929–1933
• Franklin D. Roosevelt — 1933–1945
• Harry S. Truman — 1945–1953
• Dwight D. Eisenhower — 1953–1961
• John F. Kennedy — 1961–1963
• Lyndon B. Johnson — 1963–1969
• Richard Nixon — 1969–1974
• Gerald Ford — 1974–1977
• Jimmy Carter — 1977–1981
• Ronald Reagan — 1981–1989
• George H. W. Bush — 1989–1993
• Bill Clinton — 1993–2001
• George W. Bush — 2001–2009
• Barack Obama — 2009–2017
• Donald Trump — 2017–2021
• Joe Biden — 2021–present

Want me to fix it for you?

• I can produce a clean text table (done above).
• If you want real presidential portraits embedded as thumbnails, I can fetch official portraits from reliable sources (Wikimedia Commons / White House) and assemble an HTML/table or a downloadable file — I’ll need to search the web for those images (want me to do that?).

Sorry for the confusion — AI can look convincing and still be wrong. Tell me how you want the corrected output (text-only, or table + real images) and I’ll fix it.

Click to expand...

It still thinks Biden is president.

Hans Blaster said:

It also thinks Donald Trome is still president. (When does "Biden"s term start?: 2066, 2086?)

Click to expand...

I'm starting to think the mess GPT-5 made with the American Presidents is either a joke or in the spirit of conspiracy theories a dastardly Chinese plot (DeepSeek) or one of OpenAi's other competitors to undermine the introduction of GPT-5.

When asked to provide a list and image of Australian Prime Ministers since 1945 it had no problems.

Here’s an updated table featuring Australian Prime Ministers since 1945, including their terms and external links to their official portraits housed in authoritative collections — like the Parliament House Historic Memorials Collection (HMC) and the National Portrait Gallery.

---

Australian Prime Ministers Since 1945​

#	Prime Minister	Term in Office	Official Portrait Link
14	Frank Forde	Jul 6–13, 1945	[HMC at Parliament House] (portrait by Joshua Smith (Wikimedia Commons, Melbourne Blogger))
15	Ben Chifley	1945–1949	[HMC – Archibald D. Colquhoun] (Wikimedia Commons, Australian Geographic)
16	Robert Menzies	1949–1966	[HMC – Ivor Hele] (Wikimedia Commons, Australian Geographic)
17	Harold Holt	1966–1967	[HMC – William Edwin Pidgeon] (Australian Geographic, Melbourne Blogger)
18	John McEwen (interim)	1967–1968	[HMC – Frederick William Leist] (Australian Geographic, Melbourne Blogger)
19	John Gorton	1968–1971	[National Portrait Gallery – photo by Nathan Kelly] (National Portrait Gallery)
20	William McMahon	1971–1972	Not available in HMC (no stable link found)
21	Gough Whitlam	1972–1975	[Purchased Archibald Prize portrait by Clifton Pugh] (Australian Parliament House, ArtsHub Australia)
22	Malcolm Fraser	1975–1983	Available in HMC (commissioned) (Australian Parliament House, ArtsHub Australia)
23	Bob Hawke	1983–1991	Available in HMC (commissioned) (Australian Parliament House, ArtsHub Australia)
24	Paul Keating	1991–1996	Available in HMC (commissioned) (Australian Parliament House, ArtsHub Australia)
25	John Howard	1996–2007	Portrait acquired by National Gallery in 1999 (National Portrait Gallery)
26	Kevin Rudd	2007–2010 & 2013	[Ralph Heimans portrait unveiled 2023] (9News)
27	Julia Gillard	2010–2013	[National Portrait Gallery – photo by Peter Brew-Bevan] (National Portrait Gallery)
28	Tony Abbott	2013–2015	Not yet commissioned (as of 2022) (ArtsHub Australia, 9News)
29	Malcolm Turnbull	2015–2018	Not yet commissioned (ArtsHub Australia)
30	Scott Morrison	2018–2022	Not yet commissioned (ArtsHub Australia)
31	Anthony Albanese	2022–present	Portrait likely forthcoming (not yet commissioned) (ArtsHub Australia)

---

Sources & Context​

• Historic Memorials Collection (HMC): This is the official collection housed in Parliament House since 1911, commissioning portraits of all Prime Ministers after their term ends (notably Gorton’s photo is in the National Portrait Gallery instead) (Australian Parliament House, ArtsHub Australia).
• National Portrait Gallery: Includes commissioned works or donations like photographs of Julia Gillard and John Gorton, and earlier acquisitions or exhibitions such as John Howard’s portrait and others from the primed exhibition (National Portrait Gallery).
• Delayed commissions: Portraits of recent Prime Ministers (Abbott, Turnbull, Morrison, Albanese) have not yet been commissioned or added to the HMC as of the latest updates through 2022 (ArtsHub Australia, 9News).

---

Would you like me to compile these into a downloadable HTML or document, similar to previous work, with these external links embedded for each portrait?

Click to expand...

This is not even a decent challenge for any LLM, when GPT-5 comes back on line I'll give it one of my work challenges.

Hans Blaster said:

Perhaps, but I doubt it. I suspect it is people using the wrong tool. An "AI" isn't needed for this task at all.

Click to expand...

Unless it was a specific test for GPT-5 to create portraits from scratch which led to the gibberish, otherwise it would have proceeded with the method used for Australian Prime Ministers which any LLM should be able to achieve.

Hans Blaster said:

Was you "prompt" something about a "table of PMs w/ links to official portraits"?

Click to expand...

No prompting from me, in fact it was GPT-5 which asked if it could use links to official portraits.
The next post is to supply GPT-5 with a research level problem which required a novel solution.

Here is the case study for the GPT-5 challenge.

First the background which was not supplied to GPT-5.

Automotive engines are susceptible to corrosion, one area in particular is the water outlet spigot to which a radiator hose is attached. The standard explanation at the time was crevice corrosion which is the result of a poorly fitted hose resulting in coolant and environmental contamination between the hose and spigot and the consumption of oxygen during the early stages of corrosion. Corrosion is accelerated when the oxygen is depleted forming an aeration or electrochemical cell composed of an oxygen depleted spigot and an oxygen rich outlet housing.

Crevice corrosion was never a satisfactory mechanism for failure as the water inlet spigot never exhibited corrosion while testing of field return parts did not detect the presence of coolant additives in the corrosion residue. The corrosion residue was also concentrated directly below the position of the hose clamp not evenly distributed if crevice corrosion was the cause.

This led to the development of a hypothesis, corrosion was caused by an electrochemical cell reaction between the outlet spigot and hose clamp. The clamp was made of stainless steel and the spigot of aluminium. Stainless steel has a lower electrode potential than aluminium making aluminium anodic in the cell reaction and corroding preferentially.
This explained why corrosion never occurred on the inlet spigot as it was made of stainless steel like the hose clamp and an electrochemical cell could not form.
The presence of an electrochemical cell was verified by measuring a voltage over the radiator clamp and outlet housing both in situ in the engine and on field return parts using a multimeter.
An accelerated corrosion test reproduced the corrosion problem by applying a 12V potential over the clamp as the cathode and the outer aluminium housing as the anode for 72 hrs.

Radiator hoses are heavily filled with carbon black to impart physical properties such as hardness and tensile strength but are also electrically conductive.
The countermeasures proposed to the engine design department in Japan (Toyota) was to use a high electrically resistant radiator hose or a powder coated hose clamp to electrically isolate the clamp both of which passed the accelerated corrosion test.
Japan surprisingly chose a third option to isolate the spigot by powder coating which was technically the most risky option as imperfections in the coating could lead to pitting corrosion.

Now for the GPT-5 test.
I gave it minimal information.

Case history: Toyota’s experienced corrosion of a spigot on the engine housing. Over the spigot was a radiator hose carrying engine coolant. Analysis of the corrosion pattern revealed the corrosion was pronounced directly under hose clip and not spread over the spigot where the hose contacted it. Consider mechanisms for corrosion and possible countermeasures.

To which it gave a detailed response.

Great case — interesting pattern. Below I’ll list the most likely corrosion mechanisms that would produce localized attack beneath the hose clip (but not across the whole hose-contact area), then give diagnostic checks you can run and practical countermeasures ranked by ease/impact.

Likely mechanisms (ranked)​

1. Crevice / oxygen-concentration cell (most likely).
   A clamp compresses the hose and creates a narrow crevice at the hose/clamp edge where electrolyte (coolant, road salt, condensation) can collect and oxygen is depleted. The metal under the clamp becomes the anode (oxygen-poor) and corrodes rapidly, while the free surfaces stay relatively passive/cathodic.
2. Galvanic corrosion from dissimilar metals.
   If the clamp or hose ferrule is a more noble metal (e.g., stainless) than the spigot material (e.g., aluminium or cast aluminium alloy), a galvanic couple will drive the spigot (less noble) to corrode where electrical contact and electrolyte exist — the clamp region is where the electrical contact + electrolyte coincide.
3. Local breakdown of protective coating/oxide.
   If the spigot has an oxide or coating, the clamp’s concentrated pressure or abrasion during installation may damage that protective layer only under the clamp, exposing bare metal to corrosive fluid and producing local attack.
4. Coolant wicking/leak and chemical attack concentrated at clamp.
   A tiny leak at the clamp edge or a pinhole in the hose can let coolant seep/wick under the clamp; trapped coolant (possibly containing chlorides or low pH) accelerates corrosion at that spot.
5. Stray/ensuing electrical current (less common).
   Stray DC current or poor grounding paths can cause anodic attack in localized areas. This is rarer but worth checking if other vehicles show electrical corrosion or if there are aftermarket electrical mods.
6. Stress corrosion or fretting (unlikely here).
   Fretting requires movement; hose contact area is usually static. Stress corrosion is less likely unless the alloy is susceptible and there’s sustained tensile stress.
   

Diagnostic checks (what to inspect / measure)​

• Remove hose and clamp, clean and visually inspect: note the exact corrosion morphology (pitting, uniform, intergranular).
• Identify spigot material (aluminium alloy, cast iron, steel, plated?). Identify clamp material (stainless, plated steel, zinc, aluminium).
• Analyze corrosion products if available (simple test: Fe vs Al particles; lab: EDX on a sample) — tells you which metal is corroding.
• Check coolant chemistry: pH, chloride content, corrosion inhibitor concentration (old coolant often becomes acidic or depleted).
• Check for damage to protective coating/oxide under clamp.
• Measure for stray current (multimeter between spigot and chassis ground) if electrical corrosion suspected.
• Check hose inner surface for perforations or wicking that could feed electrolyte into the crevice.

Practical countermeasures (immediate → design)​

Immediate / low cost (do these first)

1. Replace clamp with a rubber-lined or plastic-coated clamp (or use a worm clamp with an integrated rubber sleeve). The insulating layer prevents direct metal-to-metal galvanic coupling and reduces crevice formation.
2. Use the correct clamp material or isolate it — if spigot is aluminium, avoid bare stainless biting directly onto it; use aluminium clamp or clamp with insulating lining. If spigot is steel/iron, stainless clamp is usually fine.
3. Clean and remove corrosion, apply corrosion inhibitor / protective coating. Wire-brush the affected area, neutralize salts, then coat with a suitable high-temperature corrosion inhibitor or grease (e.g., dielectric grease or an anti-corrosion paste) before installing new hose/clamp.
4. Replace hose if compressed/damaged. Ensure hose inner surface isn’t degraded and the hose dimension matches the spigot so no micro-gaps remain.
   

Medium term / engineering controls
5. Use a clamp that spreads load / avoids point contact. A band clamp with a wider saddle or an Oetiker ear clamp with a rubber-lined sleeve spreads the compressive force and reduces crevice depth.
6. Apply an insulating sleeve between hose and spigot. A thin high-temperature rubber or polymer sleeve under the hose (or integrated rubber seat on clamp) prevents metal-metal contact and blocks galvanic current paths.
7. Surface treatment of spigot. If compatible with service temperatures and fluids, apply a durable coating (anodize for aluminium, epoxy/paint for steel, or conversion coating) or a sacrificial zinc layer if appropriate. Ensure the coating can handle clamp compression (some coatings flake under high pressure).
8. Correct clamp torque and installation practice. Over-torquing can gouge coatings and create crevices; under-torquing lets fluid migrate. Use a calibrated torque or recommended tightening practice.


Longer term / systemic fixes
9. Material compatibility review. Choose spigot and clamp materials to minimize galvanic potential (refer to galvanic series). If spigot is aluminium, prefer aluminium or insulated clamps. If spigot is iron/steel, stainless is commonly OK.
10. Coolant maintenance. Maintain correct inhibitor concentration, pH and low chloride. Replace coolant on schedule. Corrosion inhibitors protect alloys from crevice and general corrosion.
11. Redesign geometry if recurring. Modify spigot geometry so clamp sits on a flat, slightly raised boss with a built-in rubber seat or relocate clamp away from critical cast/flange edges to avoid trapping fluid.

Quick decision checklist you can follow now​

1. Remove clamp + hose and inspect. Identify base metal and clamp metal.
2. Clean corrosion and photograph pattern. If aluminium substrate shows corrosion and clamp is stainless, suspect galvanic + crevice.
3. Replace clamp with a lined / insulated clamp and fit a new hose. Use proper torque.
4. Clean and coat the spigot with a high-temp corrosion inhibitor / dielectric grease before refit.
5. Check coolant chemistry and replace/flush if degraded.

Notes on material compatibility (short)​

• Aluminium spigots + stainless clamps = risk of galvanic attack on aluminium if electrolyte is present. Either electrically isolate (rubber lining) or use aluminium-compatible clamp.
• Carbon steel spigots + stainless clamps usually result in the steel corroding (steel less noble); sacrificial coatings or stainless are OK depending on environment.
  

If you want, I can:​

• Provide a short materials compatibility chart for common spigot/clamp combos.
• Draft a step-by-step inspection and repair procedure (torques, recommended products for inhibitor/grease, part numbers for lined clamps).
• Suggest laboratory tests or sample analysis wording if you plan to send a sample for metallurgical analysis.
  

Would you like the materials compatibility chart / a step-by-step repair checklist next?

Click to expand...

The likely mechanisms listed by GPT-5 were considered in the investigation accept mechanism (2) where GPT-5 seems to have ignored there is radiator hose between the clamp and hose and there is no physical contact between the clamp, spigot and coolant in its account of galvanic corrosion. Its conclusion for galvanic corrosion is correct including the countermeasure for isolating the clamp or suggesting the Japanese method of isolating the spigot but ignoring the role of the hose it came up with the right answers for the wrong reasons.

Since mechanism (2) being diplomatic is roughly the correct answer I asked GPT-5 where did it come up with this answer, as a LLM did it find a source?

Good question — thanks for checking. Short answer: I didn’t rely on any private case file or earlier m-specific data for mechanism (2); I inferred it from standard corrosion science and common automotive materials/practices. Here’s how that inference works (plain-language, not hidden chain-of-thought):

1. Basic principle used
   • Galvanic corrosion is a well-established electrochemical phenomenon: when two dissimilar metals electrically contact each other in the presence of an electrolyte, the less-noble metal becomes anodic and corrodes preferentially. That’s a textbook cause of localized attack where two metals meet under wet conditions.
2. Why that fits your observed pattern
   • Your observation: corrosion concentrated directly under the hose clip and not across the entire hose contact area.
   • A hose clamp is a small, discrete metal contact that often provides both good electrical contact and a place for coolant or road splash to collect. If the clamp is a different (more noble) metal than the spigot, the clamp area is exactly where a galvanic cell would form and localize corrosion — matching the pattern you described.
3. Practical context that makes this plausible in cars
   • Common automotive practice: clamps are frequently stainless steel or plated steel; spigots can be aluminium, cast aluminium alloy, or plated steel. Stainless on aluminium is a known risky pairing in chloride-containing environments (road salt, degraded coolant).
   • The clamp also compresses the hose and can create crevices, so galvanic and crevice mechanisms can act together — which I noted as separate but compatible possibilities.
4. How I suggested confirming it (diagnostics)
   • Identify metals (visual, magnet test, oxide color, or lab EDX).
   • Check corrosion products (iron oxides vs aluminium hydroxides).
   • Build the galvanic-case evidence: clamp = more noble metal, electrolyte present, clear metal-to-metal contact at the corroded spot.
5. No hidden data used
   • I applied general materials-science rules plus typical automotive component choices; I did not pull a prior-specific file about this Toyota or specific part. If you want, I can search and cite automotive corrosion studies or OEM service bulletins that report similar clamp-edge corrosion.

If you’d like citations (papers, OEM service bulletins, or galvanic series charts) I can fetch recent sources and examples — want me to look them up?

Click to expand...

Despite the omission of the hose in its role for galvanic corrosion, I suspect if I had GPT-5 at the time the identification of the failure mode would have come sooner than later.