Reading Between the Dots: Decoding Hidden Computation across Filler Tokens · HackerPapers discussions

Reading Between the Dots: Decoding Hidden Computation across Filler Tokens — mentioned by E-Reverance

Sat, 29 Aug 2026 01:23:12 GMT

Neither do LLMs https://arxiv.org/pdf/2607.03502

Paper and discussion history. A mention is not an endorsement.

Reading Between the Dots: Decoding Hidden Computation across Filler Tokens — mentioned by dnautics

Fri, 14 Aug 2026 12:52:35 GMT

> to conclude if it is thinking or not I would want to look inside its head https://arxiv.org/abs/2607.03502 a non-thinking token model (just "completion") can answer one-step questions but generally not multistep questions. however, if you append [n] of a single token (e.g. period, space), it is able to use the activations in the higher layers of the blank tokens as a "scratchpad" to seemingly work through the complex question through "causal token time" and deliver a correct answer

Paper and discussion history. A mention is not an endorsement.

Reading Between the Dots: Decoding Hidden Computation across Filler Tokens — HN submission

Sun, 19 Jul 2026 09:13:19 GMT

Reading Between the Dots: Decoding Hidden Computation Across Filler Tokens

2 points; 2 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Reading Between the Dots: Decoding Hidden Computation across Filler Tokens — mentioned by kristianp

Sat, 18 Jul 2026 21:01:16 GMT

https://xcancel.com/kaleybrauer/status/2078185882926846044 Points to study at https://arxiv.org/abs/2607.03502

Paper and discussion history. A mention is not an endorsement.

Reading Between the Dots: Decoding Hidden Computation across Filler Tokens — HN submission

Sat, 18 Jul 2026 20:59:44 GMT

Reading Between the Dots: Decoding Hidden Computation Across Filler Tokens

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Reading Between the Dots: Decoding Hidden Computation across Filler Tokens — HN submission

Sat, 18 Jul 2026 12:19:46 GMT

Reading Between the Dots: Decoding Hidden Computation Across Filler Tokens

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Reading Between the Dots: Decoding Hidden Computation across Filler Tokens — HN submission

Sat, 18 Jul 2026 02:49:07 GMT

Reading Between Dots: Decoding Hidden Computation Across Filler Tokens (ICML'26)

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.