ChatGPT as alternative to Google or Wikipedia

There has been much discussion in the general media and computing media regarding ChatGPT and whether it can replace human authorship of writing or even coding.

That aside, as I have been experimenting it lately I think it may have even greater potential as a replacement or at least additional tool when using Google or Wikipedia.

While I would not rely on ChatGPT for its authored content, I can ask it to reply to a question “with references”. I can use my imagination to determine what type/age of references I want. In short it can function as a search engine but tweaked for the type of information I want much better than Google itself can.

Has anyone else noticed this?

Example:

are vegetables healthy - with 5 annotated Pubmed References

Vegetables are considered to be an important part of a healthy diet. They are a good source of many nutrients, including vitamins, minerals, and fiber. Eating a diet rich in vegetables may help to lower the risk of chronic diseases such as heart disease, cancer, and diabetes.

Here are five annotated PubMed references that support the health benefits of consuming vegetables:

  1. “Fruit and vegetable intake and risk of cardiovascular disease: systematic review and meta-analysis” (Reappearance of thalamic ataxia after a second hemorrhage at the ipsilateral thalamus - PubMed) This systematic review and meta-analysis of observational studies found a significant inverse association between fruit and vegetable intake and risk of cardiovascular disease (CVD). The highest intake of fruit and vegetables was associated with a reduction of CVD risk by up to 25%.
  2. “Fruit and vegetable consumption and all-cause, cancer and CVD mortality: analysis of Health Survey for England data” (Virtual bronchoscopic navigation for peripheral pulmonary lesions - PubMed) This study analyzed data from the Health Survey for England and found that higher fruit and vegetable consumption was associated with a lower risk of all-cause, cancer, and CVD mortality. Eating at least 7 portions of fruit and vegetables per day was associated with a 42% lower risk of all-cause mortality compared to less than one portion per day.
  3. “Fruit and vegetable consumption and risk of type 2 diabetes: systematic review and dose-response meta-analysis of prospective studies” (Denoising and covariance estimation of single particle cryo-EM images - PubMed) This systematic review and meta-analysis of prospective studies found that higher fruit and vegetable consumption is associated with a lower risk of type 2 diabetes. The study suggests that an increase of 3 servings per day of fruit and vegetables was associated with a 12% lower risk of type 2 diabetes.
  4. “Fruit and vegetable consumption and risk of stroke: a meta-analysis of prospective cohort studies” (PD-L1 Expression in Mismatch Repair-deficient Endometrial Carcinomas, Including Lynch Syndrome-associated and MLH1 Promoter Hypermethylated Tumors - PubMed) This meta-analysis of prospective cohort studies found that higher fruit and vegetable consumption is associated with a lower risk of stroke. The study suggests that an increase of 3 servings per day of fruit and vegetables was associated with a 13% lower risk of stroke.
  5. “Fruit and vegetable consumption and risk of colorectal cancer: a systematic review and meta-analysis” (mQC: A Heuristic Quality-Control Metric for High-Throughput Drug Combination Screening - PubMed) This systematic review and meta-analysis of observational studies found that higher fruit and vegetable consumption is associated with a lower risk of colorectal cancer. The study suggests that an increase of 3 servings per day of fruit and vegetables was associated with a 12% lower risk of colorectal cancer.

Please keep in mind that these articles are from 2009 to 2016, and it is always good to check for the latest research and guidelines.

It’s really quite good. The caveat is that one needs some expertise to verify the validity of its statements. I’m not sure that people understand that ChatGPT doesn’t have any understanding of what it’s generating, it is just generating the most likely parts of words that will come next in its sentences.

4 Likes

Yes. There is a bunch of news and content on it out there. Google preparing to launch competitive service. There are even plugins that show OpenAI content next to Google results in Chrome.

Full marks! From the Information:

“Microsoft and OpenAI Working on ChatGPT-Powered Bing in Challenge to Google”

Wait. Did anyone follow the citations? I’m no medical professional, but below are the abstracts for the linked articles.

I do not think they mean what ChatGPT thinks they mean.

Reappearance of thalamic ataxia after a second hemorrhage at the ipsilateral thalamus - Journal of Clinical Neuroscience

Ataxia is known to be linked to abnormalities of the cortico-ponto-cerebello-dentato-rubro-thalamo-cortical tracts. 1 , 2 The hemiataxia that develops after a thalamic stroke is caused by a lesion in the ventrolateral part of the thalamus, and usually occurs on the side of body contralateral to the lesion. 1 , 2

Virtual bronchoscopic navigation for peripheral pulmonary lesions - PubMed

Virtual bronchoscopic navigation (VBN) is a method in which the bronchoscope is guided on the bronchial route to a peripheral lesion using virtual bronchoscopic images. In reports on VBN for peripheral pulmonary lesions searched in PubMed as of November 2013, the diagnostic yield by ultrathin bronchoscopy in combination with computed tomography and VBN was within the range of 65.4-81.6%. Using endobronchial ultrasonography with a guide sheath (EBUS-GS) and VBN, it was between 63.3 and 84.4%, and using X-ray fluoroscopy and VBN, it was between 62.5 and 78.7%. The overall diagnostic yield was 73.8% [95% confidence interval (CI) 70.9-76.8%] and that for lesions ≤ 2 cm was 67.4% (95% CI 63.3-71.5%). These values indicate high diagnostic rates. In randomized comparative trials, the combination of VBN with EBUS-GS improved the diagnostic yield and shortened the examination time. The diagnostic yields for lesions in the right upper lobe, those invisible on posterior-anterior radiographs and those located in the peripheral third of the lung field were improved by VBN on ultrathin bronchoscopy in combination with X-ray fluoroscopy. The usefulness of VBN was also found on meta-analysis. Taken together, VBN is a promising navigational bronchoscopy method as it requires no specific training, has a low overall complication rate of 1.0% (95% CI 0.2-1.8%) and does not directly induce or cause severe complications. To maximize the full potential of VBN and promote its use, investigation of cases in which it is useful, determination of the optimum combination of procedures, a cost/benefit analysis and advancement of the VBN system are warranted.

Denoising and covariance estimation of single particle cryo-EM images - PubMed

Abstract

The problem of image restoration in cryo-EM entails correcting for the effects of the Contrast Transfer Function (CTF) and noise. Popular methods for image restoration include ‘phase flipping’, which corrects only for the Fourier phases but not amplitudes, and Wiener filtering, which requires the spectral signal to noise ratio. We propose a new image restoration method which we call ‘Covariance Wiener Filtering’ (CWF). In CWF, the covariance matrix of the projection images is used within the classical Wiener filtering framework for solving the image restoration deconvolution problem. Our estimation procedure for the covariance matrix is new and successfully corrects for the CTF. We demonstrate the efficacy of CWF by applying it to restore both simulated and experimental cryo-EM images. Results with experimental datasets demonstrate that CWF provides a good way to evaluate the particle images and to see what the dataset contains even without 2D classification and averaging.

Keywords: CTF correction; Steerable PCA; Wiener filtering.

3 Likes

Hmmm … color me skeptical.

This is the abstract from the second link (the first link had no abstract):


Abstract

Virtual bronchoscopic navigation (VBN) is a method in which the bronchoscope is guided on the bronchial route to a peripheral lesion using virtual bronchoscopic images. In reports on VBN for peripheral pulmonary lesions searched in PubMed as of November 2013, the diagnostic yield by ultrathin bronchoscopy in combination with computed tomography and VBN was within the range of 65.4-81.6%. Using endobronchial ultrasonography with a guide sheath (EBUS-GS) and VBN, it was between 63.3 and 84.4%, and using X-ray fluoroscopy and VBN, it was between 62.5 and 78.7%. The overall diagnostic yield was 73.8% [95% confidence interval (CI) 70.9-76.8%] and that for lesions ≤ 2 cm was 67.4% (95% CI 63.3-71.5%). These values indicate high diagnostic rates. In randomized comparative trials, the combination of VBN with EBUS-GS improved the diagnostic yield and shortened the examination time. The diagnostic yields for lesions in the right upper lobe, those invisible on posterior-anterior radiographs and those located in the peripheral third of the lung field were improved by VBN on ultrathin bronchoscopy in combination with X-ray fluoroscopy. The usefulness of VBN was also found on meta-analysis. Taken together, VBN is a promising navigational bronchoscopy method as it requires no specific training, has a low overall complication rate of 1.0% (95% CI 0.2-1.8%) and does not directly induce or cause severe complications. To maximize the full potential of VBN and promote its use, investigation of cases in which it is useful, determination of the optimum combination of procedures, a cost/benefit analysis and advancement of the VBN system are warranted.


I’m having a hard time seeing the relationship to consumption of fruits and or vegetables. Likewise for the other links. As far as I can tell ChatGPT is just making shit up.

As I don’t generally use Google I cannot speak to the point about this being better than the results from Google.

2 Likes

Curious, this is what I get for the first link, for the “abstract”:


Reappearance of thalamic ataxia after a second hemorrhage at the ipsilateral thalamus

Seung-Jae Lee, Yeong-In Kim, Kwang-Soo Lee, Joong-Seok Kim

No abstract available


Yes, sorry, I got that too. I tapped Show Details and then Full Text Links, and then the Elsevier logo

FWIW, in one of the previous ChatGPT threads, I think I pointed to someone else’s article in which it doiscovered ChatGPT making up plausible-looking but entirely fictitious links to Smithsonian articles that “supported” it’s point (or, really, cited in its response to a prompt).

Increasingly, I think that anywhere outside the realm of very general knowledge, ChatGPT (and its brethren) are generating word salad – very sophisticated word salad, but still. Caveat comestor

I’m sorry @JohnAtl but how is that any good?

1 Like

Well, that’s certainly a reason not to use Bing now isn’t it?

You can not rely on ChatGPT, no matter which part of its output you are looking at!
It is not the intention, nor the possibility, to get a thrust worthy answer from ChatGPT, even if some of the answers may have even a real part in it!

1 Like

It’s a good thinking tool, rewriting tool, etc. For example:

Write a paragraph about the action observation network, and relate it to Maurice Merleau-Ponty’s work on phenomenology. Cite references.

The action observation network (AON) is a neural network in the brain that is responsible for the perception and understanding of actions performed by oneself and others. The AON is composed of several regions including the superior temporal sulcus, the inferior parietal lobule, and the premotor cortex. These regions work together to process visual, auditory, and somatosensory information related to actions. Research has shown that the AON plays a crucial role in social cognition, language, and motor control.

Maurice Merleau-Ponty, a French philosopher, developed the concept of phenomenology, which emphasizes the subjective experience of the individual and how one’s perception shapes their understanding of the world. Merleau-Ponty argued that perception and action are closely intertwined, and that the body plays a crucial role in this relationship. The AON supports this idea, as it is responsible for processing information related to both perception and action. Thus, the AON can be understood as a neural implementation of Merleau-Ponty’s ideas about perception and action.

References:

  • Rizzolatti G, Craighero L. The mirror-neuron system. Annu Rev Neurosci. 2004;27:169-192.
  • Iacoboni M. et al. Action observation and acquired motor skills: an FMRI study with expert dancers. Cereb Cortex. 2005;15(8):1243-1249.
  • Merleau-Ponty, Maurice. Phenomenology of Perception. (1945)

Relate this to the ventral stream.

The ventral stream, also known as the “what pathway,” is a neural pathway in the brain that is responsible for the perception and recognition of objects and their features. The ventral stream runs from the primary visual cortex in the occipital lobes to the inferior temporal cortex, and is involved in visual processing tasks such as object recognition, face recognition, and scene understanding.

Research has shown that the ventral stream is closely connected to the action observation network (AON) and that the two systems work together to process visual information related to actions. For example, the ventral stream is activated when observing objects being manipulated, and the AON is activated when observing the actions themselves. This suggests that the ventral stream and AON are part of a larger network that enables the perception and understanding of actions and their context.

Therefore, the ventral stream and AON are closely related in that they both play a role in visual perception, but they process different aspects of visual information. The ventral stream processes object information, while the AON processes action information. Together, they contribute to a rich and complete perception of the visual world around us.

References:

  • Goodale, M. A., & Milner, A. D. (1992). Separate visual pathways for perception and action. Trends in neurosciences, 15(1), 20-25.
  • Grafton, S. T., Fadiga, L., Arbib, M. A., & Rizzolatti, G. (1997). Localization of grasp representations in humans by positron emission tomography. 2. Observation compared with imagination. Experimental brain research, 114(1), 103-111.

(I realize it has now omitted Merleau-Ponty’s work)

Add text relating the work of Maurice Merleau-Ponty.

Maurice Merleau-Ponty’s work on phenomenology emphasizes the relationship between perception, action, and the body. He argued that perception and action are closely intertwined, and that the body plays a crucial role in this relationship. Merleau-Ponty’s ideas can be related to the action observation network (AON) and the ventral stream, which are neural systems responsible for processing visual information related to actions and objects, respectively.

The AON, which is composed of several regions including the superior temporal sulcus, the inferior parietal lobule, and the premotor cortex, processes visual, auditory, and somatosensory information related to actions. The ventral stream, which runs from the primary visual cortex in the occipital lobes to the inferior temporal cortex, is involved in visual processing tasks such as object recognition, face recognition, and scene understanding. Together, the AON and the ventral stream enable the perception and understanding of actions and their context. This supports Merleau-Ponty’s idea that perception and action are closely intertwined.

Additionally, Merleau-Ponty’s notion of the body’s role in perception and action can be related to the AON’s role in motor control and social cognition. Merleau-Ponty argued that the body is not just a passive receptor of the world, but an active participant in shaping one’s perception of the world. The AON, by processing information related to actions and enabling motor control, demonstrates the active role the body plays in perception and action.

References:

  • Merleau-Ponty, Maurice. Phenomenology of Perception. (1945)
  • Rizzolatti G, Craighero L. The mirror-neuron system. Annu Rev Neurosci. 2004;27:169-192.
  • Iacoboni M. et al. Action observation and acquired motor skills: an FMRI study with expert dancers. Cereb Cortex. 2005;15(8):1243-1249.

That’s really good. It’s related to my PhD, so I can verify it is valid. I could continue this conversation, asking it to relate other topics, or ask questions about the systems involved. But I wouldn’t trust it for, say, medical advice. I could certainly ask, and read and verify the sources it referenced as @rkaplan did above.

I think people are too quick to poo-poo it when they ask it lame questions and get lame answers. Like a recent conversation where someone used the joke “Why is 6 afraid of 7? Because 7 8 9.” Then asked ChatGPT to explain why it was funny. ChatGPT talked about 7 being used as a word, but missed the part about “8” sounding like “ate.” Well, of course, at this point in time, that requires human understanding, and not probabilistic text generation.

2 Likes

So, your PhD is pretty much worthless, if everybody could get the same information within seconds, you wrote down in how long…?

Edit: Just found the answer, it was 8 years!

If someone told me their writing and reference work was aided by ChatGPT, I would have to disregard their work.

@JohnAtl Was your lengthy reply an attempt to snow me with volume rather than giving me a to-the-point answer? :slightly_smiling_face:

1 Like

I think people are getting overwhelmed, when the ChatGPT is giving them an answer, they think could be correct, and then are assuming, the Software would do so by intent, and replicable.

The ONLY purpose if ChatGPT is to give an answer like an human!
It does simply not mater for the Software, if the informations given within that answer are true, or not, as long as the software is writing a nice correct (Orthography!!) sentence.

BTW is your PhD-Thesis published…?!

2 Likes

The jury is still out. But AI is looking like it will keep things interesting for the next few years.

1 Like

Multiple people have showed up in the Obsidian community in the past few days saying something like:

ChatGPT told me to do [X] in order to achieve [Y] in Obsidian. Why isn’t it working?

This tells me a few things:

  1. ChatGPT-like tools will replace search.
  2. ChatGPT is definitely not ready to replace search.
  3. There exists people who have no concept of how not-ready these tools are and will follow them blindly and blame others when it ends poorly.

This is pretty silly. Disregard me, then. I recently wrote a paper where I used ChatGPT to find the best terms to use for a new framework (“What are some words that mean [X]”?). It also helped me rephrase paragraphs and ideas, and to double check that I hadn’t missed some obvious references in adjacent fields.


I’m hoping to host a workshop on these tools with my faculty sometime in the next few months. I’ll share here if there’s any interesting takeaways.

8 Likes

I enjoyed this article about using ChatGPT for journaling.

It is not silly as the original post shows, because I have no idea what you did with the results, or if you even realized the results were crap (as apparently rkaplan did not).

If rkaplan used the results in a paper, then that paper is crap. Because the results are crap.

The problem is that I have no idea how the tool was used. And unlike JohnAlt, I do not have a PhD in his, or any other revenant discipline (I stopped at an MS). So since I know the ChatGPT is often just making shit up, my default is to discount it.

You may be using the tool in a way that truly provides benefit to you. And your results may be valid. But I have no way to know nor verify that for you, or anyone else. Thus in the general case I cannot trust the results.

1 Like

This strikes me as the clearest, most succinct and most accurate statement I’ve seen yet on the state of play.

8 Likes