Getting GPT to work with external tools is harder than you think
We recorded 14 claims from this post. 9 were checked against evidence. 5 can’t be checked. 4 more are held. Read the post at its source.
Claims from this post (14)
In the order they appear in the post. Each quotation is the speaker’s words in this post. A label such as “Supported” or “Can’t be checked” is the claim’s current assessment; its page has the evidence and the reasons.
Assessment: SupportedSaid
Observation
“GPT-4 and similar current language-based AI programs notoriously struggle with mathematical calculations.”
Our reading: GPT-4 and similar current language-based AI programs struggle with mathematical calculations.
No horizon Evidence as of
Limitation: Results depend on operand sizes, operation mixtures, and prompting.
Assessment: Can’t be checkedSaid
Recommendation or value judgment
“That seems like a sensible idea; indeed, it is a crude version of the kind of hybrid neuro-symbolic system that we have been advocating for many years.”
Our reading: Having GPT-4 use external tools such as the Code Interpreter and Wolfram Alpha plug-ins seems a sensible idea, and it is a crude version of the hybrid neuro-symbolic system the authors have advocated for many years.
No horizon
Qualifier: seems like a sensible idea
Assessment: SupportedSaid
Causal claim
“Certainly, adding these plug-ins does improve things.”
Our reading: Adding the Code Interpreter and Wolfram Alpha plug-ins improves GPT-4's math performance.
No horizon Evidence as of
Limitation: The shared test set comprises difficult competition problems.
Assessment: UnresolvedSaid
Observation
“First, neither GPT-4 nor the plug-ins have the commonsense world knowledge and reasonable abilities needed to reliably translate a word problem into a mathematical calculation.”
Our reading: Neither GPT-4 nor its plug-ins have the commonsense knowledge and reasoning needed to reliably translate word problems into calculations.
No horizon Evidence as of
Limitation: The cited example comes from a college electromagnetism task.
Assessment: SupportedSaid
Observation
“On some problems that capable students might well find hard, the AI did perfectly; on other problems that lot of mathematically-inclined college students people would find easy, the AI failed miserably.”
Our reading: In the authors' tests, GPT-4 with plug-ins did perfectly on some problems that capable students might well find hard, and failed miserably on other problems that many mathematically inclined college students would find easy.
No horizon Evidence as of
Limitation: No student comparison group was tested, so the evidence does not establish student success rates or a measured ranking against students [e1].
Assessment: UnresolvedSaid
Observation
“In a few cases GPT-4 set up the calculations correctly and the plug-in returned the correct answer, but then GPT-4 either misunderstood or refused to accept that answer.”
Our reading: In a few test cases GPT-4 set up the calculations correctly and the plug-in returned the correct answer, but GPT-4 then either misunderstood or refused to accept that answer.
No horizon Evidence as of
Limitation: A 2024 transcript revision supplies important qualifications about the 2023 tests.
Assessment: SupportedSaid
Observation
“But, crucially, it often gets these wrong, particularly when they involve spatial reasoning beyond a cookie-cutter formula or reasoning about sequences of disparate events.”
Our reading: GPT-4 with either plug-in often gets wrong problems that a human solver would need significant thought to translate into math notation, particularly those involving spatial reasoning beyond a cookie-cutter formula or reasoning about sequences of disparate events.
No horizon Evidence as of
Limitation: The selected problems do not establish a general error rate for translation-intensive tasks or a measured comparison with human difficulty [e1].
Assessment: MixedSaid
Observation
“In many cases, a naive user might take the results, invariably presented as fact, as correct, when they are not.”
Our reading: GPT-4 with plug-ins invariably presents its results as fact, and in many cases its results are incorrect even though a naive user might take them as correct.
No horizon Evidence as of
Limitation: The mixed result depends on reading invariably literally.
Assessment: Can’t be checkedSaid
Conditional forecast
“Some of these problems may fixable. With enough resources OpenAI or Wolfram Alpha could fix some of the problems in the interfaces.”
Our reading: With enough resources, OpenAI or Wolfram Alpha could fix some of the interface problems.
No horizon
Only if: with enough resources
Assessment: Can’t be checkedSaid
Conditional forecast
“All that said, much more fundamental problems remain and, until those problems are addressed, the ability of AI to apply mathematics to real world situation will remain limited.”
Our reading: Until fundamental problems are addressed, AI's ability to apply mathematics to real-world situations will remain limited.
No horizon Evidence as of
Only if: until fundamental problems are addressed
Assessment: Can’t be checkedSaid
Causal claim
“A true understanding of elementary mathematics and its uses requires a deeper view.”
Our reading: True understanding of elementary math requires more than LLM token prediction or Wolfram-style symbol manipulation.
No horizon
Assessment: SupportedSaid
Observation
“Finding a way to get LLMs to reliably use external tools is far from trivial.”
Our reading: Getting LLMs to reliably use external tools is far from trivial.
No horizon Evidence as of
Limitation: The conclusion uses an ordinary qualitative reading of difficulty.
Assessment: UnresolvedSaid
Observation
“Thus far we have seen consistently unreliable results, from math to fact-checking.”
Our reading: So far, the authors have seen consistently unreliable results from LLMs using external tools, from math to fact-checking.
No horizon Evidence as of
Limitation: The outcome depends on retaining the explicit personal scope, 'we have seen'.
Assessment: Can’t be checkedSaid
Forecast
“We don’t claim that this problem is unsolvable, but addressing it may require major innovation.”
Our reading: Getting LLMs to reliably use external tools may require major innovation, though it is not necessarily unsolvable.
No horizon
Held from this post (4)
These records did not pass every check, so nothing about them is published. Held is not a finding.
- The two reviews reached different assessments. Our reading, not assessed: In the tests, GPT-4 often made syntax errors calling Wolfram Alpha and sometimes could not correct them.
- The two reviews reached different assessments. Our reading, not assessed: Even with plug-ins, GPT-4 cannot reliably solve word problems involving simple math or simple science and math.
- The two reviews read the claim differently
- The two reviews read the claim differently